rss_output.txt

[#<Feedjira::Parser::AtomEntry:0x00007f8c0ab73958
  @author="Drew Breunig",
  @content=
   "<p><img src=\"/img/dalle_hockney_text_factory.png\" alt=\"\" /></p>\n" +
   "\n" +
   "<h4 id=\"what-should-we-do-if-llms-arent-compatible-with-privacy-legislation\">What should we do if LLMs aren’t compatible with privacy legislation?</h4>\n" +
   "\n" +
   "<p>This week, <a href=\"https://jonathanturley.org/2023/04/06/defamed-by-chatgpt-my-own-bizarre-experience-with-artificiality-of-artificial-intelligence/\">Georgetown law professor Jonathan Turley wrote about ChatGPT repeatedly slandering him</a>:</p>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>Recently I learned that ChatGPT falsely reported on a claim of sexual harassment that was never made against me on a trip that never occurred while I was on a faculty where I never taught. ChapGPT relied on a cited Post article that was never written and quotes a statement that was never made by the newspaper. When the Washington Post investigated the false story, it learned that another AI program “Microsoft’s Bing, which is powered by GPT-4, repeated the false claim about Turley.” It appears that I have now been adjudicated by an AI jury on something that never occurred.</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>Imagine waking up one day and learning an AI is confidently claiming you’re been mired in a sexual harassment scandal. After the shock, how would you react? Turley continues:</p>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>When contacted by the Post, “Katy Asher, Senior Communications Director at Microsoft, said the company is taking steps to ensure search results are safe and accurate.” That is it and that is the problem. You can be defamed by AI and these companies merely shrug that they try to be accurate. In the meantime, their false accounts metastasize across the Internet. By the time you learn of a false story, the trail is often cold on its origins with an AI system. You are left with no clear avenue or author in seeking redress. You are left with the same question of Reagan’s Labor Secretary, Ray Donovan, who asked “Where do I go to get my reputation back?”</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>It’s a great question! And the answer, which I have been coincidentally thinking about for weeks, is far from clear. The Large Language Models (LLMs) that power ChatGPT are constructed in such a way that understanding <em>why</em> they say the things they do, let alone preventing them from saying such things, isn’t completely understood. Further, the relationship between their training data and the generated models is blurry and legally vague. <strong>LLMs – as they’re built today – may be incompatible with existing privacy regulation.</strong></p>\n" +
   "\n" +
   "<p>To understand why LLMs might not comply with regulation, we need to quickly cover how they work. And just as importantly, why they lie.</p>\n" +
   "\n" +
   "<div class=\"sidenote\">\n" +
   "\n" +
   "<h2>A Quick Note</h2>\n" +
   "\n" +
   "<p>It's particularlly hard to hold productive discussions about AI &amp; its impacts because there's so much foundation to learn, understand, and establish before we can address our chief concerns. As a result, there's a lot here. By all means, skip the upfront sections you may be familiar with. (Or, grab a drink and read the whole thing.)</p>\n" +
   "\n" +
   "<p>Sections:</p>\n" +
   "\n" +
   "<ul>\n" +
   "<li><a href=\"#how-large-language-models-work\">How Large Language Models Work</a></li>\n" +
   "<li><a href=\"#why-do-llms-make-false-claims\">Why Do LLMs Make False Claims?</a></li>\n" +
   "<li><a href=\"#how-can-you-fix-bad-training-data\">How Can You Fix Bad Training Data?</a></li>\n" +
   "<li><a href=\"#how-can-you-fix-hallucinations\">How Can You Fix Hallucinations?</a></li>\n" +
   "<li><a href=\"#so-what-is-openai-doing\">So What is OpenAI Doing?</a></li>\n" +
   "<li><a href=\"#how-can-we-do-better\">How Can We Do Better?</a></li>\n" +
   "</ul>\n" +
   "\n" +
   "</div>\n" +
   "\n" +
   "<hr />\n" +
   "\n" +
   "<h4 id=\"how-large-language-models-work\">How Large Language Models Work</h4>\n" +
   "\n" +
   "<div class=\"sidenote\">\n" +
   "\n" +
   "<p>LLMs are a collection of probabilities that perform MadLibs at a scale beyond human comprehension.</p>\n" +
   "\n" +
   "</div>\n" +
   "\n" +
   "<p>It sounds ridiculous but that’s basically it. There are plenty of excellent explanations on LLMs and how they work (<a href=\"https://arxiv.org/pdf/2212.03551.pdf\">this one</a> by Murray Shanahan might be my favorite), so we won’t be exhaustive here. But it is valuable to expand a bit.</p>\n" +
   "\n" +
   "<p>First, let’s describe how one creates an LLM:</p>\n" +
   "\n" +
   "<ol>\n" +
   "  <li><strong>Get a TON of text data:</strong> They don’t call them “Large Language Models” without reason. You’re going to need a comical amount of text. It can be typed text (from forum posts, blogs, articles, books, etc) or transcribed text (podcasts, audiobooks, or video transcriptions). It doesn’t matter, you just need a lot – we’ll explain why in a bit.</li>\n" +
   "  <li><strong>Convert the text to numbers:</strong> If we keep our data in text form during training, our processors have to deal with the words, phrases, and punctuation <em>as textual data</em>. And processors don’t natively understand text. So we translate the words into something closer to a processor’s lingua franca: numbers. We’ll walk through our giant dataset of text and build a dictionary of each word or punctuation we encounter. But rather than definitions, each word maps to a number. After we create this dictionary we can ‘tokenize’, or translate, our input data by converting each word to its associated number. A simple sentence like, “See spot run,” is encoded into something like: <code class=\"language-plaintext highlighter-rouge\">[242359, 3493, 12939]</code>\n" +
   "After the text is tokenized, the tokens are converted into “<a href=\"https://en.wikipedia.org/wiki/Word_embedding\">word embeddings</a>”. Word embeddings worth another post entirely, but in a nutshell: word embeddings are a set of coordinates that map words by <em>contextual similarity</em>. Words used in similar contexts are given embedding values that are close together (for example: “car”, “automobile”, and “bike” will be in the same region). This helps models genericize questions and contexts so that “Tell me how an apple tree grows,” will yield a similar response to, “Explain how an apple tree grows.” (Embeddings are interesting and worth checking out, if you haven’t already.) But once our words are numeric representations, we can improve our model’s map.</li>\n" +
   "  <li><strong>Build a statistical map of how tokens contextually relate to one another:</strong> Now our processors will pour over our encoded input data using a technique called ‘<a href=\"https://en.wikipedia.org/wiki/Unsupervised_learning\">unsupervised learning</a>’. Training data is continually fed into the training program and the model tries to predict the next word (or really, the next token) given a context. If it’s correct (or close), the statistical association is reinforced. If it’s incorrect, the weighting between that word and context is dialed down. This guessing and checking is the “learning” part of “machine learning.” The model is gradually improved as more and more training is performed. Ultimately, this training work builds a map of which words are most likely to follow specific patterns of other words. This ‘map’ is our LLM model.</li>\n" +
   "</ol>\n" +
   "\n" +
   "<p>At the core of ChatGPT, Bard, Bing, or LLaMA is a model like the one we ‘built’ above. Again, I’ve greatly simplified things here, but those missing details are less important than appreciating a fact I’ve underemphasized above: <em>the sheer size of every step</em>.</p>\n" +
   "\n" +
   "<p>Let’s use GPT-3 to illustrate the scope of these things:</p>\n" +
   "\n" +
   "<ul>\n" +
   "  <li><strong>How Big: How much text data was used?</strong> GPT-3 was trained on 499 billion tokens (remember, each token corresponds with a word, punctuation, chunk of a word, etc.). Sources aren’t clear on the total size of this dataset, but for comparison GPT-2 was trained on only 10 billion tokens, which equated to 40GB of <em>text</em>. The amount of data here is more than you could read in hundreds of lifetimes, if your read constantly.</li>\n" +
   "  <li><strong>How Hard: How much context was examined?</strong> When building a LLM, the amount of processing you have to do isn’t just dependent on how large your input data is. The size of the <em>context</em> you use to compute your statistical relationships matters too. The amount of tokens in your input data roughly equates to how many steps you have to take while the size of your context defines the <em>size</em> of those steps. For example, if your context size was only the preceding word every comparison only requires you store and process 2 words of data. That’s very cheap but yields a bad model. GPT-3 had a context size of <em>2,049 tokens</em>. So with every token processed, you had to store and process a significant chunk of data. (For GPT-4 this window skyrocketed, ranging between 8,192 and <em>32,768</em> tokens).</li>\n" +
   "  <li><strong>How Expensive: How much compute was used?</strong> In <a href=\"https://lambdalabs.com/blog/demystifying-gpt-3\">a technical analysis</a>, Lambdalabs estimated it would cost $4.6 million and 355 years to train GPT-3 on a single GPU. Take this number with a grain of salt, as OpenAI isn’t using a single GPU (<em>clearly</em>) and has countless optimizations going on behind the scenes. But the point to take away here is: <em>building LLMs takes a lot of time and money</em>.</li>\n" +
   "</ul>\n" +
   "\n" +
   "<p>All of this produces a model with 175 billion “parameters”. Each “parameter” is a computed statistical relationship between a token and a given context. And this is just GPT-3!</p>\n" +
   "\n" +
   "<p>The only thing that isn’t big about LLMs is the filesize of the model they output. For example, one of Meta’s <a href=\"https://ai.facebook.com/blog/large-language-model-llama-meta-ai/\">LLaMA models</a> was trained on one trillion tokens and produced a final model whose size is only 3.5GB! In a sense, LLMs are a form of <em>file compression</em>. Importantly, this file compression is <em>lossy</em>. Information is lost as we move from training datasets to models. We cannot look at a parameter in a model and understand <em>why</em> it has the value it does <em>because the informing data is not present</em>.</p>\n" +
   "\n" +
   "<p>(Sidenote: this file compression aspect of AI is a generally under-appreciated feature! Distilling giant datasets down to tiny files (a 3.5GB LLaMA model fits easily on your smartphone!) allows you to bring capabilities previously tied to big, extensive, remote servers to your local device! This will be game-changing. But I digress…)</p>\n" +
   "\n" +
   "<p>With the model built above you can now perform <em>text prediction</em>. Shanahan, in the <a href=\"https://arxiv.org/pdf/2212.03551.pdf\">paper I linked above</a>, describes this well:</p>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>Suppose we give an LLM the prompt “The first person to walk on the Moon was ”, and suppose it responds with “Neil Armstrong”. What are\n" +
   "we really asking here? In an important sense, we are not really asking who was the first person to walk on the Moon. What we are really asking\n" +
   "the model is the following question: Given the statistical distribution of words in the vast public corpus of (English) text, what words are most likely to follow the sequence “The first person to\n" +
   "walk on the Moon was ”? A good reply to this question is “Neil Armstrong”.</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>And:</p>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>Suppose you are the developer of an LLM and you prompt it with the words “After the ring was destroyed, Frodo Baggins returned to ”, to which it responds “the Shire”. What are you doing here? On one level, it seems fair to say, you might be testing the model’s knowledge of the fictional world of Tolkien’s novels. But, in an important sense, the question you are really asking … is this: Given the statistical distribution of words in the public corpus, what words are most likely to follow the sequence “After the ring was destroyed, Frodo Baggins returned to”? To which an appropriate response is “the Shire”.</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>The magic trick here, why LLMs work <em>so well</em> when all they’re doing is mostly guessing based on context, is the sheer size of their input data. It’s a scale of input beyond human comprehension and human <em>expectation</em>. The trick is not unlike a TikTok video that shows someone performing an insanely difficult and unlikely action (like a basketball trickshot) while omitting the thousands of failed attempts (this tension is a central theme in Christopher Nolan’s <a href=\"https://en.wikipedia.org/wiki/The_Prestige_(film)\">The Prestige</a>).</p>\n" +
   "\n" +
   "<p>We’ve covered a lot here, so let’s highlight the key points that matter for our privacy discussion:</p>\n" +
   "\n" +
   "<div class=\"sidenote\">\n" +
   "\n" +
   "<h2>How LLMs Work</h2>\n" +
   "\n" +
   "<ul>\n" +
   "<li>Giant amounts of text training data are used to build models.</li>\n" +
   "<li>Models express statistical relationships between words and surrounding context.</li>\n" +
   "<li>Models are reductive, distilling giant training sets into (relatively) tiny statistical models.</li>\n" +
   "<li>It takes lots of time and money to build good models.</li>\n" +
   "</ul>\n" +
   "\n" +
   "</div>\n" +
   "\n" +
   "<hr />\n" +
   "\n" +
   "<h4 id=\"why-do-llms-make-false-claims\">Why Do LLMs Make False Claims?</h4>\n" +
   "\n" +
   "<p>Let’s get back to Jonathan Turley being falsely associated with sexual harassment claims and the question of what recourse is available. <em>How can one respond to false AI claims?</em> Before we can answer this we need to build on what we’ve established above to understand <em>how and why</em> AIs lie.</p>\n" +
   "\n" +
   "<p>Unfortunately, Turley is not alone in discovering falsehoods about himself in LLMs. Noting LLM hallucinations (as they’re called) has become a bit of an internet sport.</p>\n" +
   "\n" +
   "<p>Spurred by these hallucinations, <a href=\"https://arstechnica.com/information-technology/2023/04/why-ai-chatbots-are-the-ultimate-bs-machines-and-how-people-hope-to-fix-them/\">Benj Edwards published an interesting piece at Ars Technical exploring why ChatGPT is so good at bullshit</a>. The whole piece is worth a read, but there are a few nice points relevant to us.</p>\n" +
   "\n" +
   "<p>First, Edwards <a href=\"https://arxiv.org/abs/2109.07958\">references a paper</a> authored by researchers from Oxford and OpenAI which spotted two major types of falsehoods that LLMs make. Edwards writes:</p>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>The first comes from inaccurate source material in its training data set, such as common misconceptions (e.g., “eating turkey makes you drowsy”). The second arises from making inferences about specific situations that are absent from its training material (data set); this falls under the aforementioned “hallucination” label.</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>Edwards expands on that “hallucination” category, illustrating how it can occur in ChatGPT:</p>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>When ChatGPT confabulates, it is reaching for information or analysis that is not present in its data set and filling in the blanks with plausible-sounding words. ChatGPT is especially good at making things up because of the superhuman amount of data it has to work with, and its ability to glean word context so well helps it place erroneous information seamlessly into the surrounding text.</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>These categories inform our question, as they define <em>where</em> and <em>what</em> is causing the falsehoods.</p>\n" +
   "\n" +
   "<p>The first category is <strong>bad training data creates bad models.</strong></p>\n" +
   "\n" +
   "<p>Computer science can’t get away from the principle, “<a href=\"https://en.wikipedia.org/wiki/Garbage_in,_garbage_out\">Garbage in, garbage out</a>.” If OpenAI’s training data contains falsehoods (and it does, as much of it is <em>text found on the internet</em>), their resulting models will have these erroneous associations baked in. If these falsehoods are <em>repeated</em> throughout the training data, they are further reinforced. To LLMs, <em>frequency equals fact</em>.</p>\n" +
   "\n" +
   "<p>We can see these types of errors by trying out common misconceptions in Chat-GPT. Here’s a popular error in action:</p>\n" +
   "\n" +
   "<p><img src=\"/img/seal_in_juices.png\" alt=\"Searing steak does not trap in juices.\" /></p>\n" +
   "\n" +
   "<p><a href=\"https://www.seriouseats.com/the-food-labs-top-food-myths\">Searing a crust onto a steak doesn’t trap juices</a>, but it’s a common enough misconception that ChatGPT regurgitates the sequence. <strong>Common misconceptions significantly influence models because they are common.</strong></p>\n" +
   "\n" +
   "<p>In the case of Jonathan Turley, if there were plenty of slanderous blog posts, articles, and comments online claiming he’s been accused of sexual harassment we could expect ChatGPT to echo these claims.</p>\n" +
   "\n" +
   "<p>The second category of errors is “hallucination” errors: <strong>when LLMs are asked to complete prompts that weren’t well-covered by its training data it will make up an answer based on very little data.</strong></p>\n" +
   "\n" +
   "<p>A good example of hallucinations is when an LLM cites articles or books that don’t exist, even giving page numbers as a reference. These non-existent citations don’t occur because they exist somewhere in the training data, but because the LLM is trying to complete a sequence <em>without an understanding of how words relate beyond their contextual similarity</em>. This is why it botches references and citations so spectacularly: it’s not referencing a passage in a text, <em>it’s trying to complete an expected sequence of words that make up the reference</em>. It doesn’t help that citations and references follow standardized patterns, so their specific patterns <em>frequently</em> occur and enforce those parameters in an LLM’s model.</p>\n" +
   "\n" +
   "<p>For example, Hacker News user <a href=\"https://news.ycombinator.com/user?id=msravi\">msravi</a> entered into ChatGPT the prompt, “Please provide references that deal with the mathematical properties of lists.” <a href=\"https://news.ycombinator.com/item?id=33841672\">ChatGPT answered</a>:</p>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>“Introduction to the Mathematics of Lists” by Peter J. Cameron (https://www.cambridge.org/core/books/introduction-to-the-mathematics-of-lists/3F0C6D0E69E49BF6D5CD6CA5D6A761B2)</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>“List Processing and Programming” by John F. Sowa (https://www.cs.cmu.edu/~fp/courses/lp/lectures.html)</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>“List Processing in Prolog” by David H. D. Warren (https://www.cs.purdue.edu/homes/hosking/726/html/prolog_lists.html)</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>“Introduction to the Mathematics of Lists and Trees” by Meinard Müller (https://link.springer.com/book/10.1007%2F978-3-319-05147-1)</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>“Every single one of those references and links are made up. The references don’t exist and the links appear to be cobbled together,” <a href=\"https://news.ycombinator.com/item?id=33841672\">msravi wrote</a>.</p>\n" +
   "\n" +
   "<p>Hallucinations are weird, often eerie, but they’re an expected artifact from the model building and usage process. As we’ve covered, LLMs are doing their best to guess the next word in a sequence given everything they’ve processed in their training data. If the sequence they’re given is novel they’ll just guess and toss out words with contextually similar values. This is problematic when there’s little relationship between a text’s factual meaning and the context within which it’s usually used.</p>\n" +
   "\n" +
   "<p>Once again, we’ve covered a lot here. We’re inadvertently illustrating the challenge of holding productive discussions about AI. There’s so much to learn and understand <em>before</em> we can address our chief concerns.</p>\n" +
   "\n" +
   "<p>To summarize:</p>\n" +
   "\n" +
   "<div class=\"sidenote\">\n" +
   "\n" +
   "<h2>Why LLMs Lie</h2>\n" +
   "\n" +
   "<ul>\n" +
   "<li>Models produce false outputs due to false training data or non-existent training data.</li>\n" +
   "<li>The more common a misconception or lie is, the more it will be reinforced.</li>\n" +
   "<li>'Hallucinations' occur when LLMs try to respond to sequences they haven't seen before.</li>\n" +
   "<li>When the factual meaning of a text doesn't closely relate to the context within which it's usually used, hallucinations are more likely to occur.</li>\n" +
   "</ul>\n" +
   "\n" +
   "</div>\n" +
   "\n" +
   "<hr />\n" +
   "\n" +
   "<h4 id=\"how-can-you-fix-bad-training-data\">How Can You Fix Bad Training Data?</h4>\n" +
   "\n" +
   "<p>Now that we know the types of errors that can occur, how can we fix them?</p>\n" +
   "\n" +
   "<p>For erroneous input data, the fix seems simple: delete the bad training data. This <em>would</em> be a great solution if training LLMs didn’t take tremendous time and money. Further, it’s currently impossible to know which parameters a specific bit of training data informed and by how much. The reductive process of training an LLM means we can’t easily reach in and pull out a contribution from a specific article or articles.</p>\n" +
   "\n" +
   "<p>Instead of addressing the root issue (the false training data), LLM maintainers like OpenAI rely on a technique called <a href=\"https://en.wikipedia.org/wiki/Reinforcement_learning_from_human_feedback\">Reinforcement Learning from Human Feedback</a>, or RLHF. This is a fancy way of saying continuing to train LLMs with active input from <em>people</em>, rather than data.</p>\n" +
   "\n" +
   "<p>For example, to address the erroneous sexual harassment claims against Jonathan Turley, OpenAI might pay lots of contractors to give their models prompts about Turley and provide negative feedback when any links to sexual harassment emerge.</p>\n" +
   "\n" +
   "<p>OpenAI uses RLHF to prepare its LLMs for wider audiences. Countless contractor hours are ‘correcting’ the GPTs’ outputs, discouraging toxic output or uses OpenAI would like to prohibit. OpenAI also uses RLHF to train the GPTs to understand its users better: out of the box, LLMs are sequence completion machines, <a href=\"https://openai.com/research/instruction-following#sample2\">but ChatGPT responds appropriately to questions and commands thanks to RLHF</a>.</p>\n" +
   "\n" +
   "<p>Reinforcement Learning from Human Feedback is one way we can correct an LLM’s lies caused by bad training data. But this too is costly (paying contractors is expensive!) and <em>not 100% effective</em>. OpenAI can pay for lots of contractors to spend lots of time to skew its problematic parameters, but they can’t cover every type of input. Countless “<a href=\"https://en.wikipedia.org/wiki/Prompt_engineering\">prompt engineers</a>” have been probing the edges of ChatGPT, finding contrived ways to evade its human-led coaching.</p>\n" +
   "\n" +
   "<p>And when I say contrived, I mean <em>contrived</em>. <a href=\"https://www.jailbreakchat.com/prompt/4f37a029-9dff-4862-b323-c96a5504de5d\">Here’s an example “jailbreak” prompt people are using to evade ChatGPT’s training</a>:</p>\n" +
   "\n" +
   "<p><img src=\"/img/chatgpt_larp.png\" alt=\"Asking ChatGPT to role play so it does things it's shouldn't.\" /></p>\n" +
   "\n" +
   "<p>I can’t imagine how many contractors you’d have to employ to cover this scenario!</p>\n" +
   "\n" +
   "<p>The inability of RLHF to completely squash unwanted output raises significant questions, both regarding privacy and safe harbor regulation. There’s no standard for how much counter-training a company should perform to correct an erroneous output. We lack metrics for quantifying how significantly a bad bit of training data contributes to a model or a model’s tendency for lying about a particular topic. Without these metrics, we can’t adequately define when a model is in or out of compliance.</p>\n" +
   "\n" +
   "<div class=\"sidenote\">\n" +
   "\n" +
   "<h2>How to (Try to) Fix Errors Caused By Bad Training Data</h2>\n" +
   "\n" +
   "<ul>\n" +
   "<li>It's incredibly expensive to delete the bad data and retrain the model.</li>\n" +
   "<li>There's no current, practical method for identifying and deleting the contribution from a bit of bad training data.</li>\n" +
   "<li>Unwanted output can be corrected by adding *more* training data, in the form of humans providing feedback (RLHF).</li>\n" +
   "<li>RLHF moderation isn't perfect because it can't anticipate all contexts.</li>\n" +
   "</ul>\n" +
   "\n" +
   "</div>\n" +
   "\n" +
   "<hr />\n" +
   "\n" +
   "<h4 id=\"how-can-you-fix-hallucinations\">How Can You Fix Hallucinations?</h4>\n" +
   "\n" +
   "<p>Ok, but what about hallucinations? How can we fix errors spewed by LLMs not caused by bad training data but by <em>unfamiliar prompts</em>?</p>\n" +
   "\n" +
   "<p>One way to fix this is adding <em>more</em> training data, accounting for more potential contexts. This is partially what OpenAI did with GPT-4, which has roughly 1,000 times more parameters than GPT-3. By adding more (hopefully correct!) input data, GPT-4 is less likely to be surprised by a novel prompt. This is a great <em>general</em> strategy, but might not solve one-off complaints like those of Turley. In these instances, companies like OpenAI again rely on Reinforcement Learning from Human Feedback (RLHF) to correct errors, just as they do with bad input data. Humans add more training, one prompt at a time, to moderate a specific issue.</p>\n" +
   "\n" +
   "<p>Another option to fix hallucinations is to tell LLMs to make less dramatic guesses. When an LLM doesn’t have a clear answer for a given prompt or sequence, it will select less associated tokens, introducing a bit of randomness into the response. When standing up an existing model, we can define <em>how random</em> this guess should be. We call this parameter, “temperature.”</p>\n" +
   "\n" +
   "<p>When Bing’s chatbot (née <a href=\"https://www.theverge.com/2023/2/23/23609942/microsoft-bing-sydney-chatbot-history-ai\">Sydney</a>) broke down in public, <a href=\"https://www.nytimes.com/2023/02/16/technology/bing-chatbot-transcript.html\">culminating in an attempt to break up Kevin Roose’s marriage</a>, Microsoft acted by dialing down the temperature parameter. Further, they gave users indirect control over the temperature by <a href=\"https://arstechnica.com/information-technology/2023/03/microsoft-equips-bing-chat-with-multiple-personalities-creative-balanced-precise/\">allowing them to choose</a> between “creative”, “balance”, and “precise” modes.</p>\n" +
   "\n" +
   "<p><img src=\"/img/bing_creative.jpg\" alt=\"Bing's chatbot allows selecting a conversation style, which moderates the temperature.\" /></p>\n" +
   "\n" +
   "<p>Bing’s chatbot settings present another mechanism for controlling LLM lies: user interfaces. LLMs can be accessed with more restrictive interfaces to limit the contexts given. In the image above, we can see Microsoft doing this a bit by providing preset prompts as options (“That’s too complicated. Can you suggest something simpler?”)</p>\n" +
   "\n" +
   "<p>One could take this further and remove the open-ended text box and only allow inputs via tightly constrained controls. For example, consider an app that lets you take a picture of a restaurant menu and spits out recipes for recreating each meal at home. This use case is well within GPT’s wheelhouse and <em>eliminates</em> the potential for slanderous statements in exchange for <em>greatly</em> limiting the flexibility of the tool.</p>\n" +
   "\n" +
   "<p>And that flexibility is not only valuable to people using Bing, Bard, and ChatGPT: it’s incredibly valuable to Microsoft, Google, and OpenAI. <em>Because when you use these tools you’re training them.</em> Reinforcement training by humans <em>is</em> cheaper than retraining models from scratch <em>but it’s not cheap</em>. But we train GPTs for free while playing with ChatGPT, hitting those thumbs up and down buttons, telling it something doesn’t make sense, and thanking it for a job well done.</p>\n" +
   "\n" +
   "<p>ChatGPT is a <a href=\"https://www.dbreunig.com/2016/06/23/the-business-implications-of-machine-learning.html\">Reciprocal Data Application</a>, or RDAs. As I wrote back in 2016: “[Reciprocal Data Applications] are designed to spur the creation of training data as well as deliver the products powered by the data captured. People get better apps and companies get better data.” RDAs are a new kind of network effect. More usage means a better product, which results in more usage, and so on.</p>\n" +
   "\n" +
   "<p>This is why ChatGPT is (mostly) free. This is why OpenAI pushed to be first to market, despite the risks. And this is most likely why OpenAI seems to be avoiding the privacy discussion and its implications entirely.</p>\n" +
   "\n" +
   "<div class=\"sidenote\">\n" +
   "\n" +
   "<h2>How to (Try to) Fix Errors Caused By Hallucinations</h2>\n" +
   "\n" +
   "<ul>\n" +
   "<li>We can reduce the amount LLMs guess when they're in uncharted contexts by dialing back the 'temperature'.</li>\n" +
   "<li>UIs can limit the potential for falsehoods and bad output by limiting the ways users can interact with LLMs, like removing open-ended text boxes.</li>\n" +
   "<li>But open-ended interfaces are valuable to users (who value the flexibility) and LLM owners (who value the training).</li>\n" +
   "</ul>\n" +
   "\n" +
   "</div>\n" +
   "\n" +
   "<hr />\n" +
   "\n" +
   "<h4 id=\"so-what-is-openai-doing\">So What is OpenAI Doing?</h4>\n" +
   "\n" +
   "<p>Now, finally, we can get to the crux of the issue. The reason Turley feels so powerless to address the slander emanating from ChatGPT.</p>\n" +
   "\n" +
   "<p>As we’ve detailed, there <em>are</em> mechanisms for moderating LLMs. But <em>all</em> of them are imperfect.</p>\n" +
   "\n" +
   "<p>Reinforcement Training is expensive and can’t completely eliminate bad output. Dialing down LLM’s ability to guess when it doesn’t have a strong recommendation makes them boring. Hiding LLMs behind more restrictive UIs makes them less appealing to users and restricts user feedback that continues to train the LLM. Deleting problematic training data and retraining models is <em>incredibly</em> expensive and time-consuming.</p>\n" +
   "\n" +
   "<p>OpenAI <em>knows</em> how imperfect these options are, especially anything that involves retraining from scratch. And it seems their approach to addressing this issue is <em>avoiding the problem</em>. Rather than fostering a discussion about how to proceed, their communications and documents either ignore the issue entirely or hide behind vague wording.</p>\n" +
   "\n" +
   "<p>Let’s take a look at OpenAI’s <a href=\"https://openai.com/policies/privacy-policy\">Privacy Policy</a>. To comply with laws like <a href=\"https://leginfo.legislature.ca.gov/faces/codes_displayText.xhtml?division=3.&amp;part=4.&amp;lawCode=CIV&amp;title=1.81.5\">CCPA</a> and <a href=\"https://gdpr-info.eu\">GDPR</a>, businesses are required to post a privacy policy that details the personal data they collect.</p>\n" +
   "\n" +
   "<p>In their Privacy Policy OpenAI never mentions their training data. They detail the data they collect when users sign up for and use an OpenAI account: your email, name, IP address, cookies, and such. They enumerate your rights (conditional upon where you reside), including the right to access, correct, and delete your personal information.</p>\n" +
   "\n" +
   "<p>Again: <em>training data is never mentioned</em>. It seems to me that OpenAI is avoiding the topic and focusing on <em>registration</em> data because of the challenges we detailed above.</p>\n" +
   "\n" +
   "<p><strong>Correcting or deleting personal information in their training data may require OpenAI to pay contractors to perform imperfect, reinforcement training or retrain the entire model from scratch. The legal precedents have not been established and existing regulations were not written with LLMs in mind.</strong></p>\n" +
   "\n" +
   "<p>We know OpenAI knows this because <em>they acknowledge the challenge elsewhere</em>. In a post published on April 5th, titled “<a href=\"https://openai.com/blog/our-approach-to-ai-safety\">Our Approach to AI Safety</a>,” OpenAI writes:</p>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>While some of our training data includes personal information that is available on the public internet, we want our models to learn about the world, not private individuals. So we work to remove personal information from the training dataset where feasible, fine-tune models to reject requests for personal information of private individuals, and respond to requests from individuals to delete their personal information from our systems. These steps minimize the possibility that our models might generate responses that include the personal information of private individuals.</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>Here they acknowledge that personally identifiable information (<a href=\"https://www.dol.gov/general/ppii\">PII</a>) exists in their training dataset and that they aren’t able to entirely comply with correction or deletion requests. Removal “where <em>feasible</em>” and steps that “<em>minimize</em> the possibility” that models “<em>might</em> generate responses that include personal information” (emphasis mine) is not ideal! That these acknowledgments exist in a blog post and not their Privacy Policy is even worse.</p>\n" +
   "\n" +
   "<p>It’s hard to not read the above without seeing the very <em>careful</em> yet <em>imprecise</em> language as an artifact of a key tension here: it is obvious that training data is governed by data regulations but it is <em>not</em> obvious that the LLM models themselves are governed by the same regulations.</p>\n" +
   "\n" +
   "<p>Certainly, LLM models are not 1-to-1 mappings of their training datasets. But if not, what are they? In one sense, they could be considered <em>derived abstractions</em> of the training data. In another sense, they are <em>distillations</em> of the training data. It doesn’t help that AI boosters market them as the latter when championing them but retreat to the former when defending them.</p>\n" +
   "\n" +
   "<p>We can see this tension on display elsewhere in OpenAI’s “<a href=\"https://openai.com/blog/our-approach-to-ai-safety\">Our Approach to AI Safety</a>” post:</p>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>Our large language models are trained on a broad corpus of text that includes publicly available content, licensed content, and content generated by human reviewers. We don’t use data for selling our services, advertising, or building profiles of people—we use data to make our models more helpful for people. ChatGPT, for instance, improves by further training on the conversations people have with it.</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>Sure, OpenAI doesn’t <em>directly</em> use their training data for “selling”, “advertising”, or “building profiles of people.” But according to their <a href=\"https://openai.com/policies/terms-of-use\">Terms of Use</a> they 100% allow companies using APIs powered by GPT to do <em>exactly</em> these use cases. The Terms of Use even list specific, additional standards OpenAI users need to meet if they plan on processing personal data.</p>\n" +
   "\n" +
   "<p>Reading the “<a href=\"https://openai.com/blog/our-approach-to-ai-safety\">Approach to AI Safety</a>” post and the <a href=\"https://openai.com/policies/terms-of-use\">Terms of Use</a> we are left to conclude that <strong>OpenAI does not view the actions of its models as utilization of their training data.</strong></p>\n" +
   "\n" +
   "<p>Finally, Jonathan Turley might argue that the act of training an LLM <em>is</em> “building profiles of people.” An LLM doesn’t <em>know</em> who Jonathan Turley is; it is not a database in the usual sense. But it <em>does</em> have parameters linking his name (of the tokens of his name) to contexts and tokens that are presented as knowledge.</p>\n" +
   "\n" +
   "<p>We can see what OpenAI is doing about the problem of erroneous responses in the above. In their communications and documents they minimize privacy issues or simply never mention them.</p>\n" +
   "\n" +
   "<p>I am sympathetic. As Steward Brand once said, “If you want to know where the future is being made, look for where language is being invented and lawyers are congregating.” I believe AI is reconfiguring the future (though I hate the term ‘AI’ and think it especially hurts in contexts like this.) OpenAI has been leading the field and setting the pace.</p>\n" +
   "\n" +
   "<p>By being aggressive with ChatGPT’s launch, and making it available to all, they’re reaping the benefits of constant feedback. Each conversation held with ChatGPT pushes them out further. This is an <em>entirely</em> new interface and they’re gaining experience while everybody else just tries to launch. The <a href=\"https://www.dbreunig.com/2016/06/23/the-business-implications-of-machine-learning.html#the-rise-of-reciprocal-data-applications-rdas\">Reciprocal Data Application</a> is a design with network effects, which OpenAI accrues every day they’re live and at the forefront. It’s an enviable position.</p>\n" +
   "\n" +
   "<p>In exchange for this lead, they become the <em>target</em> of criticism and concern. But they <em>must</em> lean into the tough conversations, not cede them to be hashed out and written without them. To start, they need to invest in their communication (and not more AGI handwringing, for christ sake) because false accusations against Turley are nothing compared to <a href=\"https://www.dbreunig.com/2023/02/24/we-need-ai-safewords.html\">the fraud that will bubble up shortly</a>. Even if bad actors use LLaMA, they’re going to catch flack.</p>\n" +
   "\n" +
   "<div class=\"sidenote\">\n" +
   "\n" +
   "<h2>How OpenAI is Addressing Privacy</h2>\n" +
   "\n" +
   "<ul>\n" +
   "<li>Their Privacy Policy ignores the existence of personal identifiable information (PII) in their training data.</li>\n" +
   "<li>Their own blog post seem to acknowledge their inability to totally remove PII from their training data and their model output.</li>\n" +
   "<li>The same blog post says they don't use training data for advertising or profiling, but do not prohibit these use cases in their Terms of Use.</li>\n" +
   "<li>Rather then highlighting and discussing the differences between training data and models, and openly discussing the regulatory questions, OpenAI seems to deliverately avoid this tension.</li>\n" +
   "</ul>\n" +
   "\n" +
   "</div>\n" +
   "\n" +
   "<hr />\n" +
   "\n" +
   "<h4 id=\"how-can-we-do-better\">How Can We Do Better?</h4>\n" +
   "\n" +
   "<p>To foster this industry <em>and</em> respect the people whose data is being used to build these models, all LLM owners should consider the following steps:</p>\n" +
   "\n" +
   "<ol>\n" +
   "  <li>Sponsor and invest in projects focusing on data governance and auditing capabilities for models, so we might modify them directly rather than simply adding corrective training.</li>\n" +
   "  <li>Be completely transparent regarding the training data used in these models. Don’t merely list categories of data, go further. Perhaps even make the metadata catalogs public and searchable.</li>\n" +
   "  <li>Work on defining standard metrics for quantifying the significance and surface area of output errors, and associated standards for how they should be corrected.</li>\n" +
   "  <li>Always <a href=\"https://www.dbreunig.com/2023/02/24/we-need-ai-safewords.html\">disclose AI <em>as AI</em></a> and present it as fallible. More “artificial,” less “intelligence.”</li>\n" +
   "  <li>Collaboratively establish where and how AI can or should be used in specific venues. Should AI be used to grant warrants? Select tenants? Price insurance? Triage healthcare? University admissions?</li>\n" +
   "</ol>\n" +
   "\n" +
   "<p>Collaboration is key here. Right now OpenAI is asking us to take them at their word and trust them without any established history. Outside organizations – including companies, industry groups, non-profits, think tanks, and government agencies – are already starting to kick off their own, separate conversations. Just today, <a href=\"https://www.axios.com/2023/04/11/ai-safety-rules-commerce-department-artificial-intelligence\">the US Commerce Department invited public comments to inform policy recommendations</a>. These individual entities are all trying to figure this out, separately. As the current leader, OpenAI has an opportunity to help bring these together into productive collaborations.</p>\n" +
   "\n" +
   "<p>But if AI companies cede these conversations by avoiding them, their future will be written by reactions to bad news. OpenAI and others need to invest in the boring stuff, have the hard conversations, to foster a world-changing technology. Otherwise, they risk arresting AI’s potential.</p>",
  @entry_id=
   "https://www.dbreunig.com/2023/04/10/the-privacy-question-and-open-ai",
  @links=
   ["https://www.dbreunig.com/2023/04/10/the-privacy-question-and-open-ai.html"],
  @published=2023-04-10 00:00:00 UTC,
  @raw_title="AI Lies, Privacy, &amp; OpenAI",
  @summary="",
  @title_type="html",
  @updated=2023-04-10 00:00:00 UTC,
  @url=
   "https://www.dbreunig.com/2023/04/10/the-privacy-question-and-open-ai.html">,
 #<Feedjira::Parser::AtomEntry:0x00007f8c0abd97d0
  @author="Drew Breunig",
  @content=
   "<h4 id=\"standard-commands-confirming-an-ais-presence-is-a-common-sense-starting-point-for-ai-controls\">Standard commands confirming an AI’s presence is a common sense starting point for AI controls</h4>\n" +
   "\n" +
   "<p><img src=\"/img/weights.png\" alt=\"\" /></p>\n" +
   "\n" +
   "<p>The effectiveness of <a href=\"https://arxiv.org/abs/2212.03551\">Large Language Models</a> has spurred an AI renaissance. We’re awash in models capable of generating images from mere phrases and chatbots capable of holding confident, emotional conversations. The noise of crypto subsided just as this generation of deep learning models refigured technology conversations seemingly overnight, posing BIG questions that seemed years away only months ago.</p>\n" +
   "\n" +
   "<p>People who hadn’t been following AI were whiplashed by the performance of ChatGPT, Stable Diffusion, and others. They asked questions about copyright law, the exploitation of training data, and wondered about labor regulations and worker rights if AI renders entire jobs obsolete. Others, who’ve been swimming in AI for years, set their sights <em>far</em> beyond these concerns and <a href=\"https://openai.com/blog/planning-for-agi-and-beyond/\">began to prepare for <em>true</em> artificial intelligence</a>, or <a href=\"https://en.wikipedia.org/wiki/Artificial_general_intelligence\">AGI</a>.</p>\n" +
   "\n" +
   "<p>The wide range of applications these models seem to address has boggled our minds and diluted our conversations. Only now do we seem to be stepping back and gathering our senses. We’re seeing <a href=\"https://www.nytimes.com/2023/02/16/technology/bing-chatbot-transcript.html\">these models have limits and flaws</a>. We’re starting to see that they’re not gods or monsters. Just organized bundles of statistics beyond the scale of human comprehension.</p>\n" +
   "\n" +
   "<p>In this moment I’d like to ignore the big picture questions and argue for something relatively mundane: <a href=\"https://en.wikipedia.org/wiki/Safeword\">safewords</a>.</p>\n" +
   "\n" +
   "<p>I believe fraudulent misuse of models poses an immediate risk. The ability of interactive models to generate voices, videos, and text capable of impersonating humans will become a powerful tool for social engineering <em>at scale</em>. The most common scams already focus their attention on those unfamiliar with technology (especially seniors) to manipulate access to bank and e-commerce accounts. These tactics are so effective, <a href=\"https://thediplomat.com/2022/08/inside-southeast-asias-casino-scam-archipelago/\">large market ecosystems have emerged to execute them at scale</a>, powered by human trafficking. This slavery is both what enables these scams and limits them, as the scam itself is dependent on canny humans manipulating marks one by one. New AI threatens to <em>remove</em> this limitation, scaling the scams beyond what even slavery allows and potentially leading to a diverse array of smaller-money tactics which previously hadn’t been worth the time.</p>\n" +
   "\n" +
   "<p><strong>The threat of AGI is a far-off dream. LLM-powered fraud is kicking off <em>now</em>.</strong></p>\n" +
   "\n" +
   "<p><strong>A relatively quick regulatory action we can take is requiring government-approved safewords to be built into AI models. These safewords, when input into an AI-powered interface, would confirm to users they are interacting with an AI and provide basic metadata detailing the given model.</strong></p>\n" +
   "\n" +
   "<p>I haven’t thought through all the details, but it’s worth sketching this out in hopes of kickstarting the conversation.</p>\n" +
   "\n" +
   "<p>Adding safewords to <em>base models</em> – the largest, foundational models which are tuned into custom applications (GPT is one such base model) – is an effective mechanism for adding regulation without hindering the ecosystem at large. Large base models require significant resources to build, limiting the number of parties able to develop them. Safewords added as custom rules to these models will not hinder their effectiveness for all non-fraudulent use cases.</p>\n" +
   "\n" +
   "<p>There is history and comparables here: AI safewords are akin to WHOIS and other <a href=\"https://lookup.icann.org/en\">ICANN mechanisms</a> for website accountability and transparency. Optional standards, like <a href=\"https://moz.com/learn/seo/robotstxt\">robots.txt</a>, also define ways forward and models for implementation.</p>\n" +
   "\n" +
   "<p>Requiring AI-powered interfaces and models to respond appropriately to approved safewords will not eliminate fraudulent behaviors by bad actors. But it lays the groundwork for enforcement mechanisms that governments can use to police bad actors and reduce the ease with which actors can leverage the largest, most effective models for illicit actions. Further, such requirements and enforcement should <em>not</em> hinder AI innovation. So long as a model complies with safeword requirements, leeway is granted.</p>\n" +
   "\n" +
   "<p>Finally, AI safewords provide a tangible escape hatch for users to utilize when AI models start to cross emotional lines. We humans are hardwired to engage in anthropomorphism, seeing the mark of intelligence where only hyper-scale cut-and-paste exists. By building in standard mechanisms for confirming the a model’s inhumanity, we allow users to ground themselves on occasion.</p>",
  @entry_id="https://www.dbreunig.com/2023/02/24/we-need-ai-safewords",
  @links=["https://www.dbreunig.com/2023/02/24/we-need-ai-safewords.html"],
  @published=2023-02-24 00:00:00 UTC,
  @raw_title="We Need AI Safewords",
  @summary=
   "Standard commands confirming an AI’s presence is a common sense starting point for AI controls",
  @title_type="html",
  @updated=2023-02-24 00:00:00 UTC,
  @url="https://www.dbreunig.com/2023/02/24/we-need-ai-safewords.html">,
 #<Feedjira::Parser::AtomEntry:0x00007f8c0e7253c0
  @author="Drew Breunig",
  @content=
   "<h4 id=\"a-primer-on-media-measurement-and-why-it-defines-your-world\">A Primer on Media Measurement and Why it Defines Your World</h4>\n" +
   "\n" +
   "<p><img src=\"/img/measured_dollar.jpeg\" alt=\"\" /></p>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>We sometimes treat the information industries as if they were like any other enterprise, but they are not, for their structure determines who gets heard.</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>— Tim Wu, <a href=\"https://amzn.to/2YTMMHw\">The Master Switch</a></p>\n" +
   "\n" +
   "<p>Advertising funds much of the information and entertainment in our lives. The amount of funding from ads each source receives is determined by media metrics. Media metrics help advertisers understand who is looking at their ads (if anyone) and what they’re doing after they see them.</p>\n" +
   "\n" +
   "<p>Perform well against key media metrics and your media property will thrive. Fail to deliver and your newspaper, TV station, website, podcast, or newsletter will need to find money elsewhere.</p>\n" +
   "\n" +
   "<p>Media metrics influence what information circulates because they determine which information sources gets paid. And content, like most anything else, follows the easy money.</p>\n" +
   "\n" +
   "<hr />\n" +
   "\n" +
   "<p><img src=\"/img/pressure_gauge.png\" alt=\"\" /></p>\n" +
   "\n" +
   "<h3 id=\"what-are-media-metrics\">What Are Media Metrics?</h3>\n" +
   "\n" +
   "<p>Media metrics tell advertisers if their ads are worthwhile. They answer questions, like:</p>\n" +
   "\n" +
   "<ul>\n" +
   "  <li>Are there people at the venue where I’m advertising?</li>\n" +
   "  <li>Are the people I care about at this venue?</li>\n" +
   "  <li>Are people seeing my ad?</li>\n" +
   "  <li>Are people interacting with my ad?</li>\n" +
   "  <li>Are people influenced by my ad?</li>\n" +
   "  <li>Are people taking action based on my ad?</li>\n" +
   "</ul>\n" +
   "\n" +
   "<p>Some of these questions are easy to answer. Some of them are nearly impossible. All of them are heavily contested and argued about due to the money involved.</p>\n" +
   "\n" +
   "<p>Media metrics are tools marketers use in an attempt to answer one or more of the questions above.</p>\n" +
   "\n" +
   "<p>In digital advertising, the most common media metrics are:</p>\n" +
   "\n" +
   "<ul>\n" +
   "  <li><strong>Cost Per Click (CPC):</strong> How much does the advertiser have to spend to get someone to click one of their ads?</li>\n" +
   "  <li><strong>Cost Per Action (CPA):</strong> How much does the advertiser have to spend to get someone to perform a predefined action? Often this is defined as a the audience buying something, signing up for something, or calling a phone number.</li>\n" +
   "  <li><strong>Cost Per Install (CPI):</strong> A variant of CPA. How much does the advertiser have to spend to get someone to install an app?</li>\n" +
   "  <li><strong>Cost Per Mille (CPM):</strong> How much does the advertiser have to spend for one thousand people to view their ad? (‘Mille’ just means thousand.) CPM is often defined by a target audience. For example, a marketer pays to deliver an ad to affluent Millennials.</li>\n" +
   "</ul>\n" +
   "\n" +
   "<p>There are many more metrics and new ones appear all the time. But most fail or are used by a handful of marketers for their own purposes. (Domino’s Pizza used “Cost Per Pizza Ordered” for years. They may still!)</p>\n" +
   "\n" +
   "<p>Very, very rarely a new metric will be adopted by a sufficient number of marketers and media owners to become a currency: a common metric people use to buy and sell media. A currency creates a marketplace. It defines the demand present, incentivizing content producers to create things for it.</p>\n" +
   "\n" +
   "<p>Currencies are the metrics you should be aware of and understand. They’re the metrics that shape our world.</p>\n" +
   "\n" +
   "<p><img src=\"/img/many_bills.png\" alt=\"\" /></p>\n" +
   "\n" +
   "<h3 id=\"creating-a-media-currency\">Creating a Media Currency</h3>\n" +
   "\n" +
   "<p>There’s no clear moment when a metric graduates to become a currency. It happens slowly but predictably. More and more advertisers allocate funds to buy a new metric, driving more and more media outlets sell inventory metered by this metric. The metric becomes portable, liquid. A stable, recurring source of ad spending.</p>\n" +
   "\n" +
   "<p>The creation of a media currency is a disruptive event in the marketplace. The dollars must come from somewhere. They’re either newly allocated marketing dollars (because the metric represents an entirely new type of opportunity) or shuffled from an incumbent currency (because the metric supplants existing tactics). Internal Marketing groups, media agencies, and publishers rise and fall when a new currency is crowned.</p>\n" +
   "\n" +
   "<p>Because of this, new currencies are not easily created. Incumbent industries fight their emergence.</p>\n" +
   "\n" +
   "<p>But this is how a currency is usually created:</p>\n" +
   "\n" +
   "<p><img src=\"/img/gem.png\" alt=\"Gem\" class=\"center\" /></p>\n" +
   "\n" +
   "<h4 id=\"phase-1-become-a-shiny-object\">Phase 1: Become a Shiny Object</h4>\n" +
   "\n" +
   "<p>Let’s say you create a new media property and/or social network and begin to acquire users and interest rapidly. Soon, your success can’t be ignored by media buyers. Your platform offers a unique experience, which cannot be measured effectively by existing media metrics. But that’s OK: your property is new, exciting, and hyped. It is a shiny object. Media buyers will buy ads to learn about the new format, impress their clients, and win awards.</p>\n" +
   "\n" +
   "<p>Most advertising budgets contain a slush fund reserved for new opportunities. Some advertisers (Red Bull, Dove, etc.) use these dollars to create exciting campaigns, unhindered by usual requirements, to win awards or shift the brand’s image. Others treat this fund as a “test and learn” budget, a place where new properties or ideas can be executed with the goal of learning, not explicit marketing success. Either way, these are the funds which you’ll be courting. Let’s say you attract lots of them.</p>\n" +
   "\n" +
   "<p>An influx in advertiser attention and money will challenge your new company. Suddenly, your platform will have two clients to keep happy instead of just one: your users and your advertisers. Maintaining focus on user happiness and growth while servicing advertisers will divide your already limited resources.</p>\n" +
   "\n" +
   "<p>As if that weren’t difficult enough, the platform has a limited life as a shiny object. Eventually you will cease to be novel and ad buyers will demand performance against media currencies. You won’t be new, cool, or exciting forever. YOU can maintain your coolness, extending shiny object phase (For example, Vice held this note impressively long) but coolness always fades.</p>\n" +
   "\n" +
   "<div class=\"sidenote\">\n" +
   "\n" +
   "<h1>Side Note</h1>\n" +
   "\n" +
   "<p>The shiny object crowd just hit a massive headwind due to macro economy troubles, so their population will dwindle. The fun stuff without metrics is the first to budget to get cut in tough times.</p>\n" +
   "\n" +
   "</div>\n" +
   "\n" +
   "<p><img src=\"/img/ruler.png\" alt=\"Gem\" class=\"center\" /></p>\n" +
   "\n" +
   "<h4 id=\"phase-2-find-a-metric-that-fits\">Phase 2: Find a Metric That Fits</h4>\n" +
   "\n" +
   "<p>Before your platform loses its cool, it needs to find or create a media metric that properly assesses its worth and doesn’t work against its unique experience. It then needs to convince media planners and their clients that the new metric is worth adopting.</p>\n" +
   "\n" +
   "<p>If a platform can’t create metric or can’t sell it to advertisers, media planners will evaluate the platform with existing metrics that will certainly paint it in an unflattering light.</p>\n" +
   "\n" +
   "<p>To understand this tension — how being valued by the wrong metric can corrupt your platform — let’s first look at a company that successfully sold in a new(ish) metric that aligned with its product. Let’s look at Google search.</p>\n" +
   "\n" +
   "<p><strong>Google’s Cost-Per-Click Metric Fits Their Product’s Strengths</strong></p>\n" +
   "\n" +
   "<p>When Google entered it’s shiny object phase the dominant metric in digital ad buying was cost-per-thousand ads delivered, or CPM. Media properties were paid every time they showed someone an ad, regardless of the outcome. If Google stuck with CPM and never sold in a new metric, there would be a strong incentive to show users more pages and keep them on their site longer — in short, deliver worse search results.</p>\n" +
   "\n" +
   "<p>Thankfully, Google adopted a lesser-used metric called cost-per-click (CPC) which billed the advertiser every time a user clicked on an ad, leaving Google’s site. This wasn’t difficult to sell to media buyers. The only limiting factors at the time were lack of organizational structure to manage ‘click’ campaigns and companies not knowing what to do with web traffic once they had it.</p>\n" +
   "\n" +
   "<p>CPC aligned the incentives for Google, who was financially encouraged to be the most efficient search engine. The more you used Google to search, the more opportunities Google had to serve you the right ad. This resonant cycle unlocked a deluge of dollars for Google, growing them into a giant.</p>\n" +
   "\n" +
   "<div class=\"sidenote\">\n" +
   "\n" +
   "<h1>Side Note</h1>\n" +
   "\n" +
   "<p>You can (and many do!) argue that the CPC model is causing Google to ship a worse product today. A search results page today is often half ads, requiring the user to scroll before any non-sponsored results are visible. But that’s another discussion entirely…</p>\n" +
   "\n" +
   "</div>\n" +
   "\n" +
   "<p>While Google found a metric that fit, most platforms aren’t so lucky. Usually they invent metrics they’re unable to sell to the market or forget to find a metric at all. They wake up one morning and discover they’re no longer a shiny object, are being valued with measures they can’t perform against, and there’s little runway left. If there’s no time to find and sell the right metric (or the market isn’t buying the ones you have) your window for monumental success has closed.</p>\n" +
   "\n" +
   "<p>Attempting to perform against existing currencies will undercut your platform, devalue what makes you unique, and lead you on a road to irrelevance.</p>\n" +
   "\n" +
   "<p>This is what happened to Tumblr.</p>\n" +
   "\n" +
   "<p><strong>Tumblr’s New Ad Formats Aren’t a Substitute for New Ad Metrics</strong></p>\n" +
   "\n" +
   "<p>Tumblr had a particularly strong shiny object phase. The company was a poster child for 2010s-era <a href=\"https://en.wikipedia.org/wiki/Silicon_Alley\">Silicon Alley</a>. It was a vibrant community with countless exciting projects emerging from it’s pages. It reeked cool. Brands wanted to be there. Corporations started their own Tumblogs alongside their campaigns and started to pester the company about advertising.</p>\n" +
   "\n" +
   "<p>In response to this demand, Tumblr launched Radar: a featured space on it’s homepage curated by editors highlighting popular or interesting posts. Radar was also (eventually) an ad unit. Brands could purchase a Radar ad for $25,000 for 6 million impressions. But this allocation of impressions didn’t include any earned impressions generated by people reblogging the post. So the CPM of the ad dropped dramatically if you created ads people were inclined to share.</p>\n" +
   "\n" +
   "<p><img src=\"/img/tumblr_radar.png\" alt=\"Tumblr Radar Ad\" /></p>\n" +
   "\n" +
   "<p>At first, this looks like a good model! The incentives were aligned for advertisers to create ads that felt at home on Tumblr, allowing Tumblr to generate revenue without driving away their users.</p>\n" +
   "\n" +
   "<p>But there were a few things that doomed Radar ads to fail.</p>\n" +
   "\n" +
   "<p>First, creating an ad people would reblog was hard. Tumblr users were creative and fickle, sniffing out in-authenticity with art student ease. Tumblr attempted to address this challenge by educating creative agencies and even offering their own in-house design services (most new networks had to do this do some degree, this wasn’t limited to Tumblr by any means).</p>\n" +
   "\n" +
   "<p>Second, Radar wasn’t targeted to specific users or content. At any given time there was a queue of rotating Radar posts, with some percentage being ads. Posts rotated through users’ Radar, fulfilling the impression quota of the ads, without consideration for who the users were. Advertisers are generally not a fan of this, though for the high priced shiny object advertising Tumblr was engaging with at the time, this was fine for a bit.</p>\n" +
   "\n" +
   "<p>But more importantly, Radar was an new ad unit not a new ad metric. Sure, CPM is a currency…but it is not a performance metric. CPM measures how many people you reach not the efficacy of the ads themselves. Radar was an ad format well suited to the shiny object phase (when ad agencies don’t care about metrics so long as they’re winning awards, learning, and/or making their client look cool).</p>\n" +
   "\n" +
   "<p>Ads at Tumblr quickly expanded beyond Radar into ‘Promoted Posts’, where advertisers paid to insert their Tumblr posts into users’ dashboards. These were priced similarly, with the same CPM metric costs being diluted if people shared good ads.</p>\n" +
   "\n" +
   "<p>But by this point, the shine was coming off Tumblr. And while they had new ad units that were well aligned with what users wanted from the platform, they nothing to show in terms of new ad metrics. Advertisers were now asking for proof the ads worked from both Tumblr and their agencies executing the buys. And Tumblr only had their diminishing novelty, Likes, and Reblogs — all of which were not denominating their ads.</p>\n" +
   "\n" +
   "<p>Around this time, Tumblr sold to Yahoo. Tumblr’s young, mobile audience checked the strategic boxes Marissa Mayer needed to check to re-frame Yahoo to the market, justifying a $1.1 billion price tag despite a pittance of revenues.</p>\n" +
   "\n" +
   "<p>After the acquisition, Tumblr finally created a metric by shifting to selling ads priced by ‘engagement’ — advertisers paid if their ads were Liked or Reblogged — but the market wasn’t having it. Too little, too late. Their shiny object status was truly gone (nothing like an acquisition to do that to you!) and they hadn’t created and sold a metric to the market.</p>\n" +
   "\n" +
   "<p>I knew decline was locked in for Tumblr when I saw this on my dashboard:</p>\n" +
   "\n" +
   "<p><img src=\"/img/tumblr_iab.jpeg\" alt=\"Tumblr's IAB Ad\" /></p>\n" +
   "\n" +
   "<p><strong>This is what capitulation looks like.</strong> Rather than a sponsored Tumblr post, this Victoria’s Secret ad is an IAB ad unit: an industry-standard digital display advertisement. The same ad format you’ll see on any other run of the mill webpage. Tumblr shoe-horned this into their dashboard with a quick blue frame, allowing them to adopt the ad units and metrics of the industry at large. They gave up creating a format and metric suited to their site and simply plugged into the broader, programmatic web. Sure, this gave them access to more money in the short term, but it guaranteed their unique platform would be valued in the same manner as any other blog, newspaper, or whatever. A race to the bottom had truly began.</p>\n" +
   "\n" +
   "<p>Tumblr failed to create a metric that suited their unique platform before they lost their shine, denying the market an effective way to measure the true value of their platform. They (smartly!) sold before this problem was fully apparent, and eventually capitulated to existing metrics, built for other sites, to easily monetize the platform on the way down.</p>\n" +
   "\n" +
   "<p>Most recently, Tumblr was sold to Wordpress for $3 million.</p>\n" +
   "\n" +
   "<p><img src=\"/img/crown.png\" alt=\"Crown\" class=\"center\" /></p>\n" +
   "\n" +
   "<h4 id=\"phase-3-crown-a-new-currency\">Phase 3: Crown a New Currency</h4>\n" +
   "\n" +
   "<p>If you can find a metric that aligns with your platform’s function and sell it to brands and ad agencies, you’re well on your way to something special. The flywheel is in motion. If you keep generating demand for your metric, at some point other platforms will take notice and adopt your metric to get a slice of the pie. Demand rises, then supply rises, and pretty soon you’ve got yourself a <strong>currency</strong> and the marketplace that comes with it. Waves of cash will engulf your platform as more advertising budgets allocate a line item for your metric.</p>\n" +
   "\n" +
   "<p>Sometimes this happens early in a platform’s lifetime. Google’s CPC adoption certainly did. Other times it takes time and context changes to create the moment ripe for a new metric. This is what happened with Facebook, whose business truly took off once their mobile-install business took off, driven by the booming smartphone install base, a Cambrian era of mobile app businesses, and their cost-per-install (CPI) metric.</p>\n" +
   "\n" +
   "<div class=\"sidenote\">\n" +
   "\n" +
   "<h1>Side Note</h1>\n" +
   "\n" +
   "<p>Facebook had several failed pushes for new metrics prior to their CPI-fueled growth period. Notable was their eGRP effort, in partnership with Nielsen. GRP (gross rating points), created and manged by Nielsen, is the currency nearly all TV ads are denominated in. Billions of ad dollars of demand are out their buying GRP. Facebook, lacking a killer metric, partnered with Nielsen to translate this metric to the Newsfeed. If successful, they could have sold ads to GRP budgets, diverting a slice of TV ad spend to their platform. However, this was wishful thinking and never resonated with the market for multiple reasons. (Twitter tried this slight of hand as well. TV budgets are huge and super enticing…)</p>\n" +
   "\n" +
   "<p>I believe this failure was actually good for Facebook. Borrowing an existing metrics not designed for their platform wouldn’t represent their unique value as well, as we saw with Tumblr. By bumbling along until they landed on CPI (which was partially due to them failing to spot mobile as the Next Big Thing!), Facebook didn’t capitulate to the value denominations of others.</p>\n" +
   "\n" +
   "</div>\n" +
   "\n" +
   "<p>This is the golden path of monumental platforms. They start by generating enough buzz to become a shiny object then use these funds to keep the company going while they try to establish a new metric that suits the platform. If they succeed, they create a currency and a marketplace, and are awash in ad dollars.</p>\n" +
   "\n" +
   "<p>But while they bask in treasure and glory, a fourth phase begins…</p>\n" +
   "\n" +
   "<p><img src=\"/img/optimize.png\" alt=\"Optimize\" class=\"center\" /></p>\n" +
   "\n" +
   "<p>Creating a currency defines demand in the marketplace. A currency highlights the existence of giant pools of money ready to buy anyone who can deliver against your metric. In short order, new companies will emerge built for the currency and existing companies will develop new ad units or features to get in on the funds.</p>\n" +
   "\n" +
   "<p>Up until now, everything we’ve covered relates to the emergence of currencies and new media platforms. Once currencies are established, the exploits emerge. This is when the currency starts to affect our media environment and how we perceive the world.</p>\n" +
   "\n" +
   "<p>After Google turned cost-per-click (CPC) into a currency, an entire industry emerged. Media owners adopted the metric and started to optimize their sites to get more clicks. Data-driven targeting, that worked across sites (so as to compete with the pristine signal Google had from user searches), blossomed into an entire sector. Search agencies become a thing, with armies of ad managers tweaking campaigns to get the lowest price per click. Botnets emerged, visiting sites and clicking on ads in increasingly convoluted fashions to evade anti-spam efforts. Everyone — ad buyers, ad sellers, tool providers — started gaming the metric. Fueled by easy money, denominated in CPC, the complexity and manipulation rose.</p>\n" +
   "\n" +
   "<p>Media currencies are a spectacular example of <a href=\"https://en.wikipedia.org/wiki/Goodhart%27s_law\">Goodhart’s Law</a>: <strong>“When a measure becomes a target, it ceases to be a good measure.”</strong></p>\n" +
   "\n" +
   "<p>This would be a curiosity to most — a framework for understanding why some popular platforms thrive and others languish — if the gaming of these metrics didn’t significantly color the information to which we’re exposed.</p>\n" +
   "\n" +
   "<p>Many online have a passing understanding of the “clickbait” concept, how content is published to perform against metrics advertisers will pay for. But the subtleties of other currencies are not fully appreciated. Metrics define demand, which define the ad formats created, which define the content being made.</p>\n" +
   "\n" +
   "<p>Perhaps my favorite recent example of this effect is on YouTube.</p>\n" +
   "\n" +
   "<p>To increase video views — one of the denominating currencies for YouTube ads — YouTube introduced mid-roll ads, or commercials in the middle of videos. YouTube only gets credit for video views if a viewer watches 10 seconds of it. Mid-roll ads generally check this box more effectively, as the viewer is already hooked on the content it’s inserted within when the ad runs. Contrast this with Pre-roll ads, which run before the content has started. Pre-rolls are more likely to drive off viewers within the first 10 seconds, as they’re not invested in the video much before it starts.</p>\n" +
   "\n" +
   "<p>As a result of this new format, driven by the video views currency, YouTube creators started making longer and longer videos. Videos longer than 8 minutes allow creators to insert mid-roll ads, and it only goes up from there. As a result, we’ve now entered the golden age of video essays with creators creating content no less than 10 minutes and often longer than an hour. The ad metric drove the creation of the format, which drove the creation of the content, remaking the media landscape on YouTube.</p>\n" +
   "\n" +
   "<p>As someone who enjoys longer form video essays, this is a great example of how tuned incentives — aligning currencies with the unique strengths of a platform — result in value for users, advertisers, and the platform. But not all currencies are as beneficial. And even the good ones age poorly over time, under optimization pressures from all sides.</p>\n" +
   "\n" +
   "<p>For generations on the internet, people have warned, “If you’re not paying, you are the product.” Such an adage is good to keep in mind, but I suggest we take it a bit further. <strong>We should understand how we’re being packaged and sold so that we understand how it shapes the environments defined for us.</strong></p>\n" +
   "\n" +
   "<p><img src=\"/img/cash.jpeg\" alt=\"\" /></p>",
  @entry_id="https://www.dbreunig.com/2022/06/06/why-media-metrics-matter",
  @links=["https://www.dbreunig.com/2022/06/06/why-media-metrics-matter.html"],
  @published=2022-06-06 00:00:00 UTC,
  @raw_title="Why Media Metrics Matter",
  @summary="A Primer on Media Measurement and Why it Defines Your World",
  @title_type="html",
  @updated=2022-06-06 00:00:00 UTC,
  @url="https://www.dbreunig.com/2022/06/06/why-media-metrics-matter.html">,
 #<Feedjira::Parser::AtomEntry:0x00007f8c097c9268
  @author="Drew Breunig",
  @content=
   "<h4 id=\"a-common-set-of-trials-faced-all-ambitious-social-networks-until-this-month\">A common set of trials faced all ambitious social networks… Until this month.</h4>\n" +
   "\n" +
   "<!-- ![](/img/1*O4yjonM7MvPjk8-saX1G1w.png) -->\n" +
   "\n" +
   "<p>Some of these hurdles just got higher and moved closer together.As Facebook faces a crisis in confidence — having lost the trust of a wide swath of users, advertisers, investors, employees, and (most worryingly) regulators — pundits have attempted to make sense of the matter by holding the company up to historical precedents. Comparisons to AOL have always been popular (<a href=\"https://kottke.org/07/06/facebook-is-the-new-aol\">here’s Kottke pairing the two over a decade ago)</a> but <a href=\"https://www.economist.com/business/2018/11/24/facebook-should-heed-the-lessons-of-internet-history\">now we’re seeing Yahoo! in the mix</a>, which is never a good sign.</p>\n" +
   "\n" +
   "<p>Considering Facebook’s challenges against the ghosts of social networks past, I couldn’t help but attempt to define a framework regarding the challenges of building a social network. Is there a consistent set of challenges these networks face on their way to either global adoption or failure? While considering the trajectories of current and past social networks, a gauntlet of sorts emerged:</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*YaK-lm_x0DJLRzyI6NPoqw.jpeg\" alt=\"The Social Networking Gauntlet\" /></p>\n" +
   "\n" +
   "<p>Ambitious social networks meet each of these successive challenges. With much effort they either overcome each subsequent challenge, stall, or entirely fail out.</p>\n" +
   "\n" +
   "<p>Let’s use this framework to think through Facebook’s current situation.</p>\n" +
   "\n" +
   "<p>How has Facebook manged to make it to Challenge 9, ‘Survive Regulation’, when all others have failed? And what challenges is it facing today?</p>\n" +
   "\n" +
   "<ul>\n" +
   "  <li>Facebook grew quickly through smart go-to-market strategies. Focusing on colleges allowed it to laser-focus on a niche and upset existing offerings (MySpace, Friendster, etc. who were struggling with Challenges 4 through 7. Facebook at the time was a ‘less boring upstart’ mentioned in Challenge 7). Only after it was the de facto school social network did it allow non-student users to join, in 2006.</li>\n" +
   "  <li>The following year, Facebook’s revenue began to grow exponentially. In 2007 annual revenue was roughly $150 million. <a href=\"https://www.statista.com/statistics/277229/facebooks-annual-revenue-and-net-income/\">Three years later, annual revenue was nearly $2 billion</a>.</li>\n" +
   "  <li>With a humming revenue machine, Facebook continued to grow and push for ubiquity. But if you try to attract <em>all *people you risk becoming overly *generic</em>, leaving yourself open to threats from smaller upstarts who pick off chunks of your user base with more tailored and exciting products. This is Challenge 7: achieve ubiquity while not becoming generic, so you’re able to defend against new networks.</li>\n" +
   "  <li>Facebook viewed this challenge as two problems, each with their own solution.</li>\n" +
   "  <li>The first problem was vanquishing more interesting upstarts. Facebook’s strategy here was to simply buy anyone who became too bothersome.</li>\n" +
   "  <li>The second component of this challenge was building a product that works for everyone without being generic. Facebook’s solution here was the algorithmic feed, which automated the adaptation of their product to their users. Sure, the scaffolding *around *the content was the same for everyone, but the content itself was not.</li>\n" +
   "  <li>The first solution worked very well. The second worked excellently until it didn’t.</li>\n" +
   "  <li>With the algorithmic feed, there were as many *versions *of Facebook as there were users. So when something went wrong, there weren’t enough resources to address the issues. There were too many Facebooks to manage.</li>\n" +
   "  <li>Caught off guard, Facebook scrambled. Turning off the algorithmic feed would tank their engagement, reduce their advertising inventory, hinder their advertising performance, and leave them open to less boring competition. <a href=\"https://www.theverge.com/2019/2/25/18229714/cognizant-facebook-content-moderator-interviews-trauma-working-conditions-arizona\">So contractors with massive work forces were hired to monitor content</a>, essentially policing algorithms with countless humans. The tech and operations are catching up, but they’re not there yet.</li>\n" +
   "  <li>Now they put out fires as they arise, waiting for automated moderation to improve and to find out what Challenge 9, ‘Survive Regulation’, has in store.</li>\n" +
   "</ul>\n" +
   "\n" +
   "<p>I think this gauntlet framework works well for framing Facebook’s success. It works pretty well for Twitter, Tumblr, Instagram, Reddit, Digg, MySpace, Friendster, Yahoo!, and more. But since I first blocked out the gauntlet, a narrative has emerged that short-circuits it a bit: TikTok.</p>\n" +
   "\n" +
   "<p>TikTok jumped from Challenge 3 to Challenge 9. Prior to the Trump administrations actions, TikTok was staffing up. It was running a familiar playbook: hiring tons of ad ops and sales people to scale their business model after acquiring and engaging a large, valuable audience. Snap did precisely this several years ago, going from just over a hundred employees to 900+ in a matter of months. TikTok was on it’s way to Challenge 4 (making money without pissing off your users) until the Trump administration sprung Challenge 9 on them.</p>\n" +
   "\n" +
   "<p>And now we’re in uncharted waters.</p>\n" +
   "\n" +
   "<p>The framework above assumes you’ve built a nicely humming revenue machine <em>before</em> you’re forced to tango with existential regulation. If TikTok <a href=\"https://www.economist.com/leaders/2020/09/19/will-tiktok-survive\">signals the beginning of more aggressive digital nationalism</a>, will this roadmap continue to make financial sense? Or have Facebook and Google timed the window perfectly (completing Challenges 1–8, globally, <em>before</em> facing 9) to become the de facto winners?</p>\n" +
   "\n" +
   "<p>While this isn’t the explicit knighting of a monopolist, <a href=\"https://amzn.to/3hPAzNm\">as predicted in Tim Wu’s The Master Switch</a>, it may be functionally the same. If the US’s regulatory response to China’s own rules (<a href=\"https://money.cnn.com/2018/02/28/technology/apple-icloud-data-china/index.html\">Apple, Microsoft, Amazon, and other’s services run on state-run servers</a>, just as TikTok’s may soon enough) spurs similar reactions from other governments the addressable market for new social network entrants is radically smaller than their established competitors. (And we haven’t even touched on how this manifests within privacy regulation.)</p>\n" +
   "\n" +
   "<p>Can you start something that will become a global social network after 2020? Or will the costs needed to even take a <em>shot</em> at major markets be too large and complex to swallow for investors? Or will less and more expensive access to global markets simply prevent new networks from ever growing to challenge the established competition? I’m not sure, but one dynamic certainly has changed:</p>\n" +
   "\n" +
   "<p>Previously, rapid success meant attracting the eye of your competition’s M&amp;A team. Today, rapid success attracts competitors <em>and</em> regulators. Which makes this trial much more expensive, much earlier. The gauntlet has been short-circuited.</p>",
  @entry_id=
   "https://www.dbreunig.com/2020/09/21/the-gauntlet-growing-social-networks-face-just-got-harder",
  @links=
   ["https://www.dbreunig.com/2020/09/21/the-gauntlet-growing-social-networks-face-just-got-harder.html"],
  @published=2020-09-21 00:00:00 UTC,
  @raw_title="The Gauntlet Growing Social Networks Face Just Got Harder",
  @summary=
   "A common set of trials faced all ambitious social networks… Until this month.",
  @title_type="html",
  @updated=2020-09-21 00:00:00 UTC,
  @url=
   "https://www.dbreunig.com/2020/09/21/the-gauntlet-growing-social-networks-face-just-got-harder.html">,
 #<Feedjira::Parser::AtomEntry:0x00007f8c097e2830
  @author="Drew Breunig",
  @content=
   "<p><img src=\"/img/1*qWliIK2AGiiV2Rs3d4S6Qw.png\" alt=\"\" /></p>\n" +
   "\n" +
   "<h4 id=\"and-why-theyre-so-powerful\">And why they’re so powerful</h4>\n" +
   "\n" +
   "<p>Recently a friend asked why people have an abundance of concern for “<a href=\"https://en.wiktionary.org/wiki/porch_pirate\">porch pirates</a>.” “Is package theft truly a frequent problem?” he asked. “Or is the inconvenience so tremendous people loath to deal with it even once?”</p>\n" +
   "\n" +
   "<p>While I tend to support the latter theory, I also believe the term “porch pirate” itself is behind the surge in concern. The term is evocative, instantly <a href=\"https://en.wikipedia.org/wiki/Grok\">grokked,</a> and puts a name to a common problem previously dealt with privately. That last part is key: nearly all of us have had packages stolen at one time in our lives, but it was never more than a personal inconvenience. You might mention it in the office kitchen the next day, killing time while coffee brews, but it wasn’t something you shared.</p>\n" +
   "\n" +
   "<p>“Porch pirate” puts a name to your personal experience and instantly makes it communal. In a moment it crystallizes internal thoughts and makes them portable. It is an exemplar <em>buzzword</em>.</p>\n" +
   "\n" +
   "<p>I am a <em>massive</em> fan of the <a href=\"https://linguistlist.org/ask-ling/sapir.cfm\">Sapir-Whorf Hypothesis,</a> which suggests that the languages we speak are the framework that we use to understand the world. Meaning: language does not just describe reality, <em>it defines reality</em>. The words you know establish the boundaries of what you can understand.</p>\n" +
   "\n" +
   "<p><a href=\"https://linguistlist.org/ask-ling/sapir.cfm\">Whorf sums it up nicely</a> (emphasis mine):</p>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>We dissect nature along lines laid down by our native languages. The categories and types that we isolate from the world of phenomena we do not find there because they stare every observer in the face; on the contrary, <strong>the world is presented in a kaleidoscopic flux of impressions which has to be organized by our minds — and this means largely by the linguistic systems in our minds</strong>. We cut nature up, organize it into concepts, and ascribe significances as we do, largely because we are parties to an agreement to organize it in this way — an agreement that holds throughout our speech community and is codified in the patterns of our language. The agreement is, of course, an implicit and unstated one, but its terms are absolutely obligatory; we cannot talk at all except by subscribing to the organization and classification of data which the agreement decrees.Sapir and Whorf focused on the differences between languages and cultures, but I most frequently reference their idea on the micro level, when new terms emerge in a single language.</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>Which brings us back to <em>buzzwords</em>.</p>\n" +
   "\n" +
   "<p>Not all new words are buzzwords. Most new words are esoteric — <em>jargon</em> — describing new concepts or ideas among groups of experts forging new ground. To understand these new words <em>you</em> must be actively learning and exploring the new terrain they occupy. Jargon emerges with expertise.</p>\n" +
   "\n" +
   "<p>Buzzwords describe what you already <em>intuitively know</em>. At once they snap the ‘kaleidoscopic flux of impressions’ in your mind into form, crystallizing them instantly allowing you to both organize your knowledge and recognize you share it with other. This rapid, mental crystallization is what I call the <em>buzzword whiplash</em>. It gives buzzwords more importance and velocity, more power, than they objectively should have.</p>\n" +
   "\n" +
   "<p>The potential energy stored within your mind is released by the <em>buzzword whiplash</em>. The buzzword is perceived as important partially because of what it describes but also because of the social and emotional weight felt when the buzzword recognizes your previously wordless experiences and demonstrates that those experiences are shared.</p>\n" +
   "\n" +
   "<p>Which is why people are suddenly concerned with “porch pirates.” The term is a perfect buzzword. And the whiplash of recognition makes people more concerned than they objectively should.</p>\n" +
   "\n" +
   "<p>You too could build a new buzzword. Here’s how to:</p>\n" +
   "\n" +
   "<ol>\n" +
   "  <li>Identify a common experience everyone intuitively is aware of but has difficulty expressing concisely (or at all).</li>\n" +
   "  <li>Craft a term to encapsulate this amorphous experience in plain and evocative language.</li>\n" +
   "  <li>Seed the term among an expressive, socially connected community in a form they want to consume and share.\n" +
   "That’s it!</li>\n" +
   "</ol>\n" +
   "\n" +
   "<p>We haven’t talked about step 3 yet, but it is just as essential as the others. “Porch pirate” registers as barely a blip in Google Trends (despite describing a common experience) until <a href=\"https://www.youtube.com/watch?v=xoxhDk-hwuo\">Mark Rober built a glitter bomb trap to catch a package thief and uploaded the footage to YouTube</a>. Rober went viral <a href=\"https://trends.google.com/trends/explore?date=today%205-y&amp;geo=US&amp;q=Porch%20Pirate\">as did the term</a>:</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*0iv19RPSU0XxWfY3sTf5xw.png\" alt=\"\" /></p>\n" +
   "\n" +
   "<p><em>Rober’s video went live on December 17th 2018. That high mark is December 16–22nd, 2018.</em></p>\n" +
   "\n" +
   "<p>Buzzwords are more powerful than what they describe. By helping you better recognize your own experiences and demonstrating they are shared, they are emotional and social explosions. Often they are accidental, like “porch pirate.” Other times they have intent, like “fake news”. It is crucial we understand how buzzwords work to counter the spells they cast.</p>\n" +
   "\n" +
   "<p>When we understand how tricks work they becomes less awesome.</p>",
  @entry_id="https://www.dbreunig.com/2020/02/28/how-to-build-a-buzzword",
  @links=["https://www.dbreunig.com/2020/02/28/how-to-build-a-buzzword.html"],
  @published=2020-02-28 00:00:00 UTC,
  @raw_title="How to Build a Buzzword",
  @summary="And why they’re so powerful",
  @title_type="html",
  @updated=2020-02-28 00:00:00 UTC,
  @url="https://www.dbreunig.com/2020/02/28/how-to-build-a-buzzword.html">,
 #<Feedjira::Parser::AtomEntry:0x00007f8c097ee7c0
  @author="Drew Breunig",
  @content=
   "<p><img src=\"/img/1*ZtDmPJ8iKzUza9ZghEl2kA.jpeg\" alt=\"Paiting by Carl Grossberg\" /> <a href=\"https://en.wikipedia.org/wiki/Carl_Grossberg\" class=\"image-caption\">Painting by Carl Grossberg</a></p>\n" +
   "\n" +
   "<h4 id=\"playing-chicken-with-the-limits-of-technology\">Playing chicken with the limits of technology</h4>\n" +
   "\n" +
   "<p>Spend enough time thinking about the technology business and you’ll inevitably acquire a mental metaphor for the rhythms and cycles you observe. A specific book or presentation might snap into your head, providing scaffolding to order successes and failures you’ve observed over the years. Or — if you are prone to metaphors and storytelling — you might develop your own.</p>\n" +
   "\n" +
   "<p>I fall firmly into the latter category. To me, a good framework for thinking about the technology business is asking, “<strong>What are we waiting for?</strong>”</p>\n" +
   "\n" +
   "<p>In 2001 we were waiting for people. Today we’re waiting for robots. But hold that thought for a moment…</p>\n" +
   "\n" +
   "<p>Benedict Evens recently published his <a href=\"https://www.ben-evans.com/presentations\">annual presentation</a> on macro trends within the technology industry. Please go read it first.</p>\n" +
   "\n" +
   "<p>In the piece, Evens asks, “What happens when everyone is online?” then follows with a quote from Marc Andreessen:</p>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>Every failed idea from the dotcom bubble would work now.There are many reasons why the dotcom bubble burst, but a dominant one is not enough people were online soon enough. Companies, caught in a game of chicken, grew as if the masses were right around the corner. When they didn’t arrive (less than 10% of the population were online at the time) companies collapsed as their investments weren’t met by customers.</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>We were waiting for people, who didn’t arrive until smartphones matured because it turned out PC were too expensive and wonky for most. Once people arrived, things started picking up:</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*HMnkfOSXMGNc68Gkg5DtsQ.png\" alt=\"\" /></p>\n" +
   "\n" +
   "<p><em>Slide from <a href=\"https://www.ben-evans.com/presentations\">Benedict Evan’s deck</a>. Ugly annotations by me.</em></p>\n" +
   "\n" +
   "<p>We wait for the limits of technology to fall. Sometimes everyone waits for a accessible, sufficient mobile UI for the internet. Sometimes a sector waits for a new type of mattress that can fit in a cardboard box.</p>\n" +
   "\n" +
   "<p><em><img src=\"/img/1*68XCIzgbbKnoVpqt24zjDQ.png\" alt=\"\" /><a href=\"https://www.ben-evans.com/presentations\">ibid</a>.</em></p>\n" +
   "\n" +
   "<p>Once a limitation is removed by an innovator, competition races to the new boundary. It’s a land rush. But if you’re investing in massive growth, <em>you venture past the limit and hope it catches up before you run out of money</em>.</p>\n" +
   "\n" +
   "<p>The limits of technology can be shared headaches or niche enablers. But despite this diversity, they move in one direction on a spectrum from <em>simple problems</em> to <em>messy problems:</em></p>\n" +
   "\n" +
   "<p><img src=\"/img/1*9h544mzM_swha3oIy2VCvw.png\" alt=\"\" />Messy problems are best solved by expensive human experts. They exist beyond the capabilities of technology. The <em>job</em> of new technology is to move problems from the messy column to the simple one.</p>\n" +
   "\n" +
   "<p>Some recent examples of messy problems made simple and the results:</p>\n" +
   "\n" +
   "<ul>\n" +
   "  <li>Smartphones made the internet <em>accessible</em> to billions, allowing eCommerce to grow.</li>\n" +
   "  <li>Vacuum-packed foam packaging made mattresses <em>portable</em>, allowing for direct-to-consumer mattress businesses to grow.</li>\n" +
   "  <li>High bandwidth connectivity made high definition video <em>portable</em>, allowing for streaming services to grow.</li>\n" +
   "  <li>High volume pricing for cellphone components made smart home devices <em>cheap</em>, allowing for our IoT Cambrian age.</li>\n" +
   "  <li>Modular shipping containers made small shipments <em>portable</em> and <em>predictable</em>, enabling the build out of global commerce.\n" +
   "In my head I make a distinction between “what we’re waiting for” (the ends) and a limit of technology (the means). Many global consumer businesses were waiting for fast, cheap shipping — not the shipping container specifically.</li>\n" +
   "</ul>\n" +
   "\n" +
   "<p>So where are the current limits of technology? What’s preventing the next S Curves, as Ben Evens asks?</p>\n" +
   "\n" +
   "<p>I keep a running list of problems that resist technology in an effort to spot patterns.</p>\n" +
   "\n" +
   "<p>Sometimes the problems don’t make the headlines, like <a href=\"https://www.latimes.com/business/story/2019-12-19/boeing-spacex-spacecraft-parachutes\">building parachutes</a>:</p>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>Parachutes encounter turbulent and dynamic airflow, which is almost impossible to replicate with computers, said Erik Seedhouse, an assistant professor in spaceflight operations at Embry-Riddle Aeronautical University. Wind speeds vary at different altitudes. Atmospheric pressure changes in a hundredth of a second. The stresses on a parachute whose job is pulling out a larger parachute can be unpredictable.Parachutes are a <em>poorly understood</em> problem; their environment and performance cannot be modeled. It’s still messy.</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>Some of the problems that resist technology are comically mundane, like <a href=\"https://www.newyorker.com/magazine/2018/02/12/why-paper-jams-persist\">paper jams</a>.</p>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p><em>“Papers are not created equally,” John Viavattine, the head of the Torture Lab, said. Some stocks generate excessive friction; others swell in the humidity. (In general, winter jams are more common than summer jams.) Sheets cut from the same forty-ream roll can vary in quality. At the center of the roll, paper fibres tend to arrange themselves in an orderly matrix; nearer the edges, they become jumbled… Even the highest-quality paper can be ruined by poor “paper handling.” A half-used package of paper left to sit will grow damp and curly or dry and “tight.” Reams of paper that are thrown around or kept in stacks can develop hidden curls that lead to jams.</em>Paper is surprising, <em>unpredictable</em>, so paper jams persist. Messy.</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>But the headline problems of our era are rooted in technology’s inability to <em>perceive</em> and <em>understand</em>. Software can read articles, organize photos, recognize faces, and spot pedestrians…but not with the skill required for massive rollout and acceptance. Software can perceive simple, predictable scenes, but it has yet to master truly messy perceptive problems. This limitation of technology is why we’re still waiting for robots, or <em>AI</em>.</p>\n" +
   "\n" +
   "<p>Facebook and Google had hopes of solving moderation with AI. But the complexities of language and diversity of rule sets (Evens gets into this multicultural challenge elsewhere in his deck) resists present technologies. Companies must hire armies of moderators.</p>\n" +
   "\n" +
   "<p>Uber can turn a profit with either complete market dominance or self-driving technology — a problem that remains messy because it’s surprising and subtle. AI, the hope goes, would address these limitations and enable &gt;10x returns.</p>\n" +
   "\n" +
   "<p>Companies that bet machine and deep learning would soon provide capable AI are in a squeeze right now. But they had little choice to not take a shot at AI, since any major advancement by a competitor could instantly provide their product at a fraction of their cost. (<a href=\"https://www.sec.gov/Archives/edgar/data/1543151/000119312519103850/d647752ds1.htm\">Uber details this dilemma quite well in their S-1</a>.)</p>\n" +
   "\n" +
   "<p>The dotcom companies of 2001 played chicken with the limits of technology and their competitors, building as if their world would arrive online tomorrow. But it took the greater part of a decade before people arrived.</p>\n" +
   "\n" +
   "<p>Today, companies are waiting for robots, sacrificing researchers and capital, hoping they’ll arrive before the money runs out or the regulation arrives. So far the problem has resisted.</p>\n" +
   "\n" +
   "<p>Which brings us back to Benedict Even’s question: “What is the next S-Curve?” I suggest the following questions as good discussions to have in pursuit of an answer:</p>\n" +
   "\n" +
   "<ol>\n" +
   "  <li>Where are the limitations of technology? What problems resist being directly addressed by software or hardware at scale?</li>\n" +
   "  <li>What are we — our current industry — waiting for? Where are we caught in games of chicken: where the rewards are so disrupting, the entire industry must invest in and chase the challenge?</li>\n" +
   "  <li>What is the advancement that has yet to arrive that we hope will put a stop to our waiting? What are the paths being explored and what are the paths less traveled?</li>\n" +
   "  <li>When the above advances arrive, what new use cases will they enable that we aren’t currently waiting for? (Note: as you can see above, the limitations of technology and what we’re waiting for are often not the same. And we often take a surprising route to arrive at our ends.)</li>\n" +
   "  <li>Poorly understood problems are the hardest technical limitations to solve. What limitations do we face today that are well understood? Poorly understood?\n" +
   "I’m sure there are more (better!) questions to help think about the future. But thinking about what we’re waiting for has been fruitful for me, at the macro and micro level.</li>\n" +
   "</ol>",
  @entry_id="https://www.dbreunig.com/2020/02/03/what-are-we-waiting-for",
  @links=["https://www.dbreunig.com/2020/02/03/what-are-we-waiting-for.html"],
  @published=2020-02-03 00:00:00 UTC,
  @raw_title="What are We Waiting For?",
  @summary="Painting by Carl Grossberg",
  @title_type="html",
  @updated=2020-02-03 00:00:00 UTC,
  @url="https://www.dbreunig.com/2020/02/03/what-are-we-waiting-for.html">,
 #<Feedjira::Parser::AtomEntry:0x00007f8c09802040
  @author="Drew Breunig",
  @content=
   "<p><img src=\"/img/1*c0dseyBJ5WcN43KVgiCIkA.png\" alt=\"One of these may have helped take down Twitter.\" /></p>\n" +
   "\n" +
   "<h4 id=\"explaining-the-ddos-problem--its-origins-as-simply-as-i-can\">Explaining the DDoS Problem &amp; its Origins, As Simply As I Can</h4>\n" +
   "\n" +
   "<p>You may have noticed the <a href=\"https://hackernoon.com/tagged/internet\">Internet</a> is kind of broken today. I think this event will become a Big Deal, potentially remembered as the denial of service attack which led to regulation and more.</p>\n" +
   "\n" +
   "<p>Discussing it via iMessage, a friend who doesn’t follow Internet <a href=\"https://hackernoon.com/tagged/security\">security</a> (like, you know, normal people) asked me to explain the above in English. I don’t have time to polish it, but I could see this being helpful to others. So here’s roughly what I write:</p>\n" +
   "\n" +
   "<ol>\n" +
   "  <li>The iPhone launched, Android followed and smartphones became <em>the</em> business. In under 10 years, roughly 30% of the world now has a smartphone. The amount of money being generated to make little devices with cameras, CPUs, WiFi, gyroscopes, touchscreens, batteries, and more boggles the mind.</li>\n" +
   "  <li>The demand for smartphones drove the costs of their components <em>wayyy</em> down. There are massive buildings in China filled with bins of components, which designers can pick up on the cheap and design a new toy. This is why almost all new device categories are basically reassembled smartphone guts with a twist: action cameras, fitness trackers, drones, hover boards, webcams, and more. They’re all plentiful and affordable because smartphones are paying for their parts en masse.</li>\n" +
   "  <li>Designers, engineers, and factories in China iterate on these products constantly. <a href=\"https://www.wired.com/2015/06/the-weird-story-of-the-viral-chinese-scooter-phunkeeduck-io-hawk/\">It’s an organic environment that plays with hardware the way Silicon Valley plays with software</a>. When one conceives, designs, produces, and ships hardware fast, on shoestring budgets, one doesn’t usually take the time to test, secure, or QA much of anything. See: exploding hoverboards. This is another way you can keep costs cheap.</li>\n" +
   "  <li>Now there are hundreds of millions of these cheap devices all over the world which are comically undersecured and connected to the Internet. And none of the companies which shipped than want to (or can) fix them. (For many of them, the insecurities are baked into the firmware.)</li>\n" +
   "  <li>Botnet makers have not missed the opportunity. They’ve developed simple hacks to remotely control millions of devices–especially webcams and routers. These devices are the perfect for botnets because they’re online all the time and widely adopted.</li>\n" +
   "  <li>Recently, the attacks from these botnets have started to reach critical limits, producing distributed denial of service attacks (wherein tons of devices all hit the same website or service as fast as they can, overwhelming it) that are factors larger than attacks they’ve seen in the past.</li>\n" +
   "  <li>This leap in strength has disrupted the tit-for-tat exchange between attackers and security firms. Security god Bruce Schneier (who I’ve been cribbing from for these last few bullets. <a href=\"http://motherboard.vice.com/read/we-need-to-save-the-internet-from-the-internet-of-things\">Read him</a>!) writes, “Basically, it’s a size vs. size game. If the attackers can cobble together a fire hose of data bigger than the defender’s capability to cope with, they win. If the defenders can increase their capability in the face of attack, they win.” The massive output of the factories, subsidized by smartphone parts, connected to a global marketplace of unaware buyers has disrupted this balance in favor of the attackers.</li>\n" +
   "  <li>And it’s already creating problems as a tool against less than desirable media. Security journalist Brian Krebs’ site was hit with an attack so large Akamai (one of <em>the</em> cloud hosting providers) <a href=\"http://www.businessinsider.com/akamai-brian-krebs-ddos-attack-2016-9?op=1\">kicked him off their servers</a>. Krebs was saved by Google, who stepped in to protect Krebs free of charge. (Worryingly, <a href=\"http://www.businessinsider.com/akamai-brian-krebs-ddos-attack-2016-9?op=1\">Krebs’ site is completely down at the moment</a>, likely connected to the current attack.)</li>\n" +
   "  <li>These weapons aren’t just disruptingly powerful, they’re relatively easily available. It’s distressing to consider the scale that could be produced by state-sponsored groups. Imagine that this Internet outage hit more than the East Coast. Imagine this outage hit on Election Day. (If there is a bright side to this attack it is that it establishes some public awareness of the issue prior to it being something which could be politically construed.)</li>\n" +
   "  <li>This attack is significant enough and targets the right people (the media in NYC and politicians in DC) to garner attention from government. Because, sadly, it appears significant regulation–both requiring basic security features for devices and granting ISPs blacklisting power–is the only way to solve this challenge. Regulating who and what can connect to the Internet appears to be necessary but sets a worrying prescident.</li>\n" +
   "  <li>In short: access to the Internet and the availability of specific sites is super vulnerable, there’s no easy fix, and the solution is bad.\n" +
   "There is a certain irony in Twitter being taken down by webcams, though.</li>\n" +
   "</ol>\n" +
   "\n" +
   "<p>So who’s doing this? Well the most reasonable–and frightening–theory comes again from Bruce Schneier: <a href=\"https://www.schneier.com/blog/archives/2016/09/someone_is_lear.html\">a state-sponsored group is testing the Internet to learn how to take it down</a>. Krebs (whose site is back up!), <a href=\"https://krebsonsecurity.com/2016/10/spreading-the-ddos-disease-and-selling-the-cure/\">thinks someone is creating the disease to sell the cure</a>. Both are possible!</p>\n" +
   "\n" +
   "<p>(There are plenty of things I’d like to link to above but cannot now, due to the outage. Apologies)</p>\n" +
   "\n" +
   "<p><strong>Update:</strong> <a href=\"https://krebsonsecurity.com/2016/10/ddos-on-dyn-impacts-twitter-spotify-reddit/#more-36727\">it really looks like this is a vendetta attack</a> against Krebs and a researcher at Dyn, the company which is the main target of the attack. The attack kicked off hours after the two of them delivered a presentation highlighting one party’s bad practices.</p>\n" +
   "\n" +
   "<p><strong>Update, November 8th, 2016:</strong> It appears we’ve been saved in an unexpected– yet totally <em>fitting</em>–way. The week of the US elections, hackers pointed the traffic cannon at the websites of both Hillary Clinton and Donald Trump, but failed to take out either. Why? Well, it turns out this particular botnet is too easy to access. Following the October 21st attack, many parties signed up the same insecure devices to do their bidding. Specifically, they wanted the botnet to attack various game hosting servers. With tons of users directing the same traffic cannon at their own pet targets, the effect of the weapon became severely dilluted.</p>\n" +
   "\n" +
   "<p>In a nutshell: the Internet lives another day because too many people wanted to grief Counter-Strike and Minecraft servers. You couldn’t script a more appropriate twist.</p>",
  @entry_id=
   "https://www.dbreunig.com/2016/10/21/how-the-iphone-led-to-today-s-internet-outage",
  @links=
   ["https://www.dbreunig.com/2016/10/21/how-the-iphone-led-to-today-s-internet-outage.html"],
  @published=2016-10-21 00:00:00 UTC,
  @raw_title="How the iPhone Led to Today’s Internet Outage",
  @summary="",
  @title_type="html",
  @updated=2016-10-21 00:00:00 UTC,
  @url=
   "https://www.dbreunig.com/2016/10/21/how-the-iphone-led-to-today-s-internet-outage.html">,
 #<Feedjira::Parser::AtomEntry:0x00007f8c0e70bee8
  @author="Drew Breunig",
  @content=
   "<p><img src=\"/img/1*a5IOD4PmyaQIJvv4HtbFxA.png\" alt=\"\" /><a href=\"https://www.instagram.com/richkidsofinstagram/\" class=\"image-caption\">Via Rich Kids of Instagram</a></p>\n" +
   "\n" +
   "<h4 id=\"participatory-mass-media-changes-the-equation\">Participatory Mass Media Changes the Equation</h4>\n" +
   "\n" +
   "<p>You should really read <a href=\"http://www.economist.com/news/briefing/21708216-americas-president-writes-us-about-four-crucial-areas-unfinished-business-economic\">Obama’s essay in The Economist</a>. In it, he discusses the rise of populism and anti-globalization, the importance of economic mobility, the growing divide between rich and poor, and how we might address these challenges.</p>\n" +
   "\n" +
   "<p>Midway in, Obama makes a quick point on how digital media affects one’s perceived social status (emphasis mine):</p>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>More fundamentally, a capitalism shaped by the few and unaccountable to the many is a threat to all. Economies are more successful when we close the gap between rich and poor and growth is broadly based. A world in which 1% of humanity controls as much wealth as the other 99% will never be stable. <strong>Gaps between rich and poor are not new but just as the child in a slum can see the skyscraper nearby, technology allows anyone with a smartphone to see how the most privileged live.</strong> Expectations rise faster than governments can deliver and a pervasive sense of injustice undermines peoples’ faith in the system. Without trust, capitalism and markets cannot continue to deliver the gains they have delivered in the past centuries.The argument that media exposes the underclass to a good life beyond their reach is not new. In his book, <a href=\"http://amzn.to/2dUTN3O\">In Spite of the Gods</a>, Edward Luce argues that access to television exacerbated unrest with regard to the economic divide in India:</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>What today’s villagers and small-town dwellers in India see seductively paraded before them as they crowd around their nearest TV screens are things most of them have little chance of getting in the near future: the cars, foreign holidays, smart medical services, and electronic gadgets that dominate the TV commercials. Most of these products are not meant for them at all but are targeted at — and often by — people like Alok. Such items are beyond the reach of the majority in a country where the average per capita income in 2006 was still below $1,000. Sooner or later if you are unable to get what you are repeatedly told you should want, something has to give.Yes, Obama is correct that smartphones and social media better expose us to the lives of the rich and famous, when compared to TV. But it goes beyond more access. Digital media’s effect on our attitude regarding socioeconomic standing is doubly problematic because we share our lives in the same forum as the privileged.</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p><img src=\"/img/1*DkRVLcTOX1xPioFaoZYg1A.png\" alt=\"On your TV\" /></p>\n" +
   "\n" +
   "<p>To audiences in the 80’s, <em>Lifestyles of the Rich and Famous</em> was practically on another planet. Robin Leach performed a function similar to David Attenborough’s: both hosts guided viewers through an environment most would never witness. Then we turned off the TV, and went back to our real world.</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*smvVsAiEmD6m8LG95oiEug.png\" alt=\"In your stream\" /></p>\n" +
   "\n" +
   "<p>But with Instagram, Snapchat, and Facebook our lives, the lives of our friends, and the lives of the famous mix in the same feed. We have the same accounts, the same phones, the same cameras. Not only is there nothing to turn off to return to our real world, we must actively consider how to present our world into a feed populated with those more privileged.</p>\n" +
   "\n" +
   "<p>This active editing, they way we’ve learned to brand ourselves and our lives, has a nasty knock-on effect: the people we know who carefully edit start to appear more privileged. I have friends of similar socioeconomic status who seem to vaction more than Leach’s subjects. Most of our feeds look this way.</p>\n" +
   "\n" +
   "<p>This change — from the lavish life on TV to lavish life on your stream––will get to many. It fundamentally changes how we <a href=\"https://en.wikipedia.org/wiki/Keeping_up_with_the_Joneses\">keep up with the Joneses</a>. In the TV era, when I wasn’t in the part of the medium, one just needed a nice lawn and adequate car. Today, if we need to feed the stream, we need better vacations, better reservations, better fashion, and free time to find diverse experiences. All at a constant pace.</p>\n" +
   "\n" +
   "<p>Unquestionably, digital media is a massive net good for the world. (Even that feels like an understatement) But its effects on perceived inequality are different than any we’ve experienced. This is a wholly new challenge. One which requires new cultural tools.</p>",
  @entry_id=
   "https://www.dbreunig.com/2016/10/08/keeping-up-with-the-joneses--today",
  @links=
   ["https://www.dbreunig.com/2016/10/08/keeping-up-with-the-joneses-today.html"],
  @published=2016-10-08 00:00:00 UTC,
  @raw_title="Keeping Up with the Joneses, Today",
  @summary="Via Rich Kids of Instagram",
  @title_type="html",
  @updated=2016-10-08 00:00:00 UTC,
  @url=
   "https://www.dbreunig.com/2016/10/08/keeping-up-with-the-joneses-today.html">,
 #<Feedjira::Parser::AtomEntry:0x00007f8c0e70dcc0
  @author="Drew Breunig",
  @content=
   "<p><img src=\"/img/1*bVGPuLFrOhHN8ig7xSgrbw.jpeg\" alt=\"\" /></p>\n" +
   "\n" +
   "<h4 id=\"does-free-speech-include-distribution\">Does ‘Free Speech’ Include Distribution?</h4>\n" +
   "\n" +
   "<p>Yes, it’s a goofy dream. Yes, Congress won’t let them stop Saturday delivery, let alone spend $30 billion on a wobbly and <em>weird</em> social network. Yes, this will never happen. Yes, $30 billion could buy 90 <a href=\"https://warisboring.com/how-much-does-an-f-35-actually-cost-21f95d239398#.2mkhzsxb8\">F-35s</a> instead.</p>\n" +
   "\n" +
   "<p>But: I can’t get this idea out of my head. My mind stumbles on it every other commute. Every news item about Twitter’s sale spurs the notion. <a href=\"http://www.recode.net/2016/10/6/13183378/recode-daily-google-disney-not-bidding-on-twitter\">Google and Disney are walking away</a> leaving only Salesforce, but oh: <a href=\"https://techcrunch.com/2016/10/03/to-beef-up-in-marketing-tech-salesforce-is-buying-krux-for-340m-in-cash-and-stock-potentially-750m-overall/\">they just bought Krux</a>. Maybe there won’t be a suitor. Their market cap is down to less than $15 billion on the news. Hmm, that’s only <em>44</em> F-35s…</p>\n" +
   "\n" +
   "<p>Ok. This won’t happen. But the idea is so natural to me, so easily raised, that I feel compelled to share it with you. Perhaps doing so will expunge it from my head.</p>\n" +
   "\n" +
   "<p>Here is why I keep dreaming about Twitter being bought by the USPS:</p>\n" +
   "\n" +
   "<ul>\n" +
   "  <li>It is becoming increasingly apparent that we, the people of the United States, expect Freedom of Speech to not only protect the articulation of one’s ideas and opinions, but the distribution of these notions as well.</li>\n" +
   "  <li>This expectation is shared by various political groups in the United States, from left and right, in various contexts.</li>\n" +
   "  <li>When Reddit banned it’s most reprehensible channels, protestors cried, “Freedom of speech!” Supporters of the move chastised the protesters for expecting that a private corporation should be required to host their content.</li>\n" +
   "  <li>When Facebook suffers a “<a href=\"http://gizmodo.com/facebook-stands-by-technical-glitch-claim-says-cop-did-1783349993\">technical glitch</a>” and removes live broadcasts of police confrontations or protests, we wring our hands about freedom of speech again. When it’s not clear how trending news is determined, representatives from both parties demand clarification. All this happens despite the fact that Facebook is another private corporation which can host, or <a href=\"http://www.nytimes.com/2016/09/10/technology/facebook-vietnam-war-photo-nudity.html?_r=0\">not host</a>, whatever it wants.</li>\n" +
   "  <li>Hell, people were shouting about freedom of speech when that dude from Duck Dynasty was kicked off a TV show.</li>\n" +
   "  <li>The expectation that freedom of speech now covers some baseline of distribution is widely held among the US population. Pundits, congress people, journalists, and average citizens have come to expect information to be distributed to some degree.</li>\n" +
   "  <li>If we wish to fulfill this expectation, we have two options for ensuring this new concept of freedom of speech is guaranteed: we can make a deal with a corporation or develop the government’s ability to meet this need. Let’s look at an example for each.</li>\n" +
   "  <li>The best example for corporate negotiated information openness is AT&amp;T. In 1913, AT&amp;T struck a deal with the US government which ensured AT&amp;T would not be pursued as a monopolist, so long as AT&amp;T allowed competiting telephone companies to interconnect with their long-distance network. This, certainly, is a gross over-simplification, but everyone should just go read Tim Wu’s <a href=\"http://amzn.to/2dPZG2j\">The Master Switch</a>, which covers all of this, in context, beautifully:</li>\n" +
   "</ul>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>We sometimes treat the information industries as if they were like any other enterprise, but they are not, for their structure determines who gets heard.</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>An AT&amp;T-type deal, the dream of an idealistic monopoly, has happened many times (and a few times before). It is the default. It is what will likely happen with Facebook, Google, or whomever wins. Seriously,<a href=\"http://amzn.to/2dPZG2j\"> just go read Wu</a>.</p>\n" +
   "\n" +
   "<ul>\n" +
   "  <li>But there is another model for information distribution: the government can take up the task. This is why we created the United States Post Office. The little discussed origins of the USPS was covered recently by <a href=\"http://amzn.to/2dQ0og0\">Winifred Gallagher</a> (<a href=\"http://www.wnyc.org/story/how-post-office-created-america/\">here is an interview with a summary of her recounting</a>).</li>\n" +
   "  <li>The USPS was instigated by our founding fathers during the 2nd Continental Congress. Ben Franklin was the first postmaster general. It is explictely authorized in the Constitution. Gallagher shows in her work that the motivation for it’s creation was to unite 13 different colonies, tying the nation together. Information, it was held, was crucial to the functioning of a democracy. If information could not be distributed, the new nation could not function.</li>\n" +
   "  <li>Both the corporate deal model and the government provided model have their pros and cons. A strong argument could be made for both. What worries me, what keeps this line of thinking alive in my brain, is that we aren’t having this discussion. We yell and protest when we feel some bit of speech <em>must</em> be distributed by Facebook, Twitter, Reddit, Google, or whomever. But we never stop and think about <em>why *and</em> if* we should be relying on private companies to guarantee the distribution of our speech.</li>\n" +
   "  <li>True, there are exchanges about net neutrality (coined by our good friend, Tim Wu) and we’ve had some victories, but we’ve had nothing approaching national discussion.</li>\n" +
   "  <li>If we do not discuss this, the choice is made for us: a deal with a corporation will be struck. We will ignore these issues while Facebook and Google continue to grow until one day we realize they’re too big and important to change or break up. Then, a deal will be cut.</li>\n" +
   "  <li>(Oh look: <a href=\"https://www.washingtonpost.com/news/the-switch/wp/2016/10/06/facebook-is-talking-to-the-white-house-about-giving-you-free-internet-heres-why-that-may-be-controversial/\">Facebook is working on bringing their own version of the Internet to rural and low income users for free</a>.)</li>\n" +
   "  <li>Or: Twitter could flounder. No suitor could emerge while their market cap dwindles. At some point their price could become within the reach of a government take-over. Then digital distribution — for text, photos, video, conversations, and more — could be a public good, maintained by the public. Facebook could block whatever it wants. Google could remove search results and the stakes would be lower. The USPS could continue their mission of ensuring equal access to the distribution of information to the nation (should they care). And finally, our contemporary expectations of freedom of speech could be reflected by government guarentees.\n" +
   "Yes, this is absurd. But it’s fun to think about. How would Twitter change under the USPS, assuming they had the resources to change Twitter? Real identities would need to be present, but not public. One could change their public name as one obtains a PO Box. Advertising would still be there, subsidizing development. The direct mail industry could move online, seamlessly shifting it’s household address databases to Twitter IDs for audience targeting. What new businesses would be unexpectedly fueled by this machine?</li>\n" +
   "</ul>\n" +
   "\n" +
   "<p>It’s a goofy idea, but one that spurs thinking.</p>\n" +
   "\n" +
   "<p>Mock it if you like, but it’s much less boring than waiting quietly to strike a deal with Facebook.</p>",
  @entry_id=
   "https://www.dbreunig.com/2016/10/07/i-dream-of-the-post-office-buying-twitter",
  @links=
   ["https://www.dbreunig.com/2016/10/07/i-dream-of-the-post-office-buying-twitter.html"],
  @published=2016-10-07 00:00:00 UTC,
  @raw_title="I Dream of the Post Office Buying Twitter",
  @summary="",
  @title_type="html",
  @updated=2016-10-07 00:00:00 UTC,
  @url=
   "https://www.dbreunig.com/2016/10/07/i-dream-of-the-post-office-buying-twitter.html">,
 #<Feedjira::Parser::AtomEntry:0x00007f8c0e710c90
  @author="Drew Breunig",
  @content=
   "<p><img src=\"/img/1*8SrauHQD5aBmPFmPFJQRwA.png\" alt=\"\" /></p>\n" +
   "\n" +
   "<h4 id=\"and-theyll-succeed-where-google-stumbled\">…and they’ll succeed where Google stumbled</h4>\n" +
   "\n" +
   "<p><a href=\"https://hackernoon.com/tagged/snapchat\">Snapchat</a> announced Spectacles, their first hardware product, last Friday evening. Despite being the first hardware effort from our* *current social darlings, Evan Spiegel and company seem to be playing down the product. In <a href=\"http://www.wsj.com/articles/snapchat-releases-first-hardware-product-spectacles-1474682719\">a Wall Street Journal feature</a> on Snap, only a handful of paragraphs touch on Spectacles. One section sandbags the entire effort:</p>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>For the moment, Spectacles appears to be a bit of a lark. At a price of $129.99 and with limited distribution, it won’t be relied upon for significant immediate revenue. Spiegel refers to it as a toy, to be worn for kicks at a barbecue or an outdoor concert — Spectacles video syncs wirelessly to a smartphone, making it easily shareable. “We’re going to take a slow approach to rolling them out,” says Spiegel. “It’s about us figuring out if it fits into people’s lives and seeing how they like it.”</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>If you ignore the Karl Lagerfeld portraits of Spiegel, the whole affair is rather low key.</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*Ee5aGGnMAt5iv6i5pOeEqg.png\" alt=\"\" />And who can blame them? Launching digital glasses anywhere in California forces comparison’s to Google Glass, a device whose <a href=\"https://www.youtube.com/watch?v=9c6W4CCU9M4\">massive, life changing promises</a> <a href=\"http://www.nytimes.com/2015/02/05/style/why-google-glass-broke.html\">were never delivered</a>. Glass’ launch video was an <a href=\"https://www.youtube.com/watch?v=9c6W4CCU9M4\">augmented reality what-if</a> followed by an <a href=\"https://www.youtube.com/watch?v=D7TB8b2t3QE#t=73\">initial live demo by skydivers</a>, then <a href=\"http://www.wsj.com/video/google-glasses-at-diane-von-furstenberg-show/738BCFD3-F507-4225-855E-D7CFA2656A7F.html\">a Diane Von Furstenburg catwalk appearance</a>.</p>\n" +
   "\n" +
   "<p>Compared to Glass as promised by Google, Spiegel’s “toy” for a “barbeque or an outdoor concert” is rather mundane.</p>\n" +
   "\n" +
   "<p>But that’s OK.</p>\n" +
   "\n" +
   "<p>I think Spectacles could not only succeed, but turn out to be a watershed product for Snap. Here’s why:</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*VEAHDMdZDzHK2SS4tLNMQQ.jpeg\" alt=\"\" /></p>\n" +
   "\n" +
   "<h4 id=\"spectacles-side-step-glass-cultural-challenges\">Spectacles Side-Step Glass’ Cultural Challenges</h4>\n" +
   "\n" +
   "<p>Despite their high technical goals, Glass’ biggest challenge was cultural. Google focused so closely on the users of Glass that they forgot everyone around them, leading to the <a href=\"https://hackernoon.com/tagged/design\">design</a> of an anti-social product. I cover this in detail <a href=\"http://dbreunig.tumblr.com/post/45752835306/google-glass-is-just-like-the-segway-and\">here</a>, but in a nutshell:</p>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>Glass’ screen is visible only to it’s user and it’s camera looks out documenting everything except the user, storing content to be shared at the user’s discretion. I believe that these always on, core functions of Glass will prevent it from being welcomed in social settings. Those around the Glass users must implicitely trust the Glass wearer, for they have no idea where the Glass user’s current attention lies and cannot visually confirm whether or not they are currently being captured by Glass.</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p><strong>Google is working so hard to keep technology out of the way that they’re forgetting why it’s important to <em>see</em>technology when it’s present</strong>.This challenge was clearly considered by the team at Snap, because they made several choices which address it directly. They added a bright light to show others you’re recording. They limited video to 10 second clips, not persistent documentation. And, most importantly, they made Spectacles sunglasses.</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>Sunglasses are worn occasionally. Sunglasses aren’t worn inside. Everyone wears sunglasses. Sunglasses make sense during public, fun times. Wear sunglasses inside and you’re considered “that guy,” no onboard camera required. Choosing sunglasses specifically tightly confines Spectacles’ use cases to social acceptable contexts. Spectacles work within cultural norms rather than attempt to redefine them.</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*9kXJtGRHC72-8l9nehXL2Q.png\" alt=\"\" /></p>\n" +
   "\n" +
   "<h4 id=\"spectacles-have-job-sian-focus\">Spectacles have Job-sian Focus</h4>\n" +
   "\n" +
   "<p>Like Job’s Apple, Snap is prioritizing focus and simplicity above all.</p>\n" +
   "\n" +
   "<p>Above is <a href=\"https://spectacles.com\">Spectacles’ website</a>. Everything we need to know about Spectacles (save the design and colors, which greets the user prior) is listed here. The promise, “Specs make memories from your perspective,” is simple and clear.</p>\n" +
   "\n" +
   "<p>Compare the above to the <a href=\"http://web.archive.org/web/20011114012126/http://www.apple.com/ipod/\">website for the first iPod</a>:</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*lowBjaArnkUCWmmM-GGzJg.png\" alt=\"\" />Same succinct five points, including one clear promise: “1,000 songs in your pocket.” <a href=\"https://medium.com/@dbreunig/m-g-is-right-that-the-watch-isn-t-a-flop-95fb92cb6718#.fujaew1cn\">This clear promise or vision was made for every Apple product that wasn’t a computer</a> (Watch excepted).</p>\n" +
   "\n" +
   "<p>Sure, this seems like best practice, but nearly every tech product which attempts to invent a category ignores this advice.</p>\n" +
   "\n" +
   "<p>Glass shouldn’t have promised an augmented life, just better memories.</p>\n" +
   "\n" +
   "<p>Alexa shouldn’t have promised household AI, just music on-demand. (Sure, Alexa’s doing OK, but compare the iPod and Spectacles sites above to <a href=\"http://amzn.to/2d1XYNY\">Alexa’s homepage</a>. Very different!)</p>\n" +
   "\n" +
   "<p>Beyond shared restraint, Specs and the iPod both feel/felt like experiments. They’re odd, but inoffensive, and I bet Specs have similarly underwhelming sales. They’ll do fine, but won’t light the world on fire. As Spiegel said, “We’re going to take a slow approach to rolling them out. It’s about us figuring out if it fits into people’s lives and seeing how they like it.”</p>\n" +
   "\n" +
   "<p><br /></p>\n" +
   "\n" +
   "<h4 id=\"when-snap-finds-the-formula-spectacles-will-be-everywhere\">When Snap Finds the Formula, Spectacles will be Everywhere</h4>\n" +
   "\n" +
   "<p>Specs are $129.99. <a href=\"http://www.ray-ban.com/usa/sunglasses/wayfarer/clv\">Stock Ray-Bans are $150</a>. Specs are already cheap enough for a fashion item. The electronic components are even cheaper.</p>\n" +
   "\n" +
   "<p>We’ll have to wait for a proper teardown, but Specs probably contain Bluetooth, a dab of storage for a few shots, several chips to run the whole thing, and a cellphone camera from several years ago. (Another benefit to making them sunglasses? The camera only needs to work in brightlight.) This package is already cheap and will only get cheaper.</p>\n" +
   "\n" +
   "<p>If Snap cracks the code and figures out this can fit in the fun parts of people’s lives, expect this kit to propagate everywhere. I wouldn’t be surprised if the salesperson at LensCrafters asks you if you’d like your sunglasses with Specs, right after asking about anti-glare coating.</p>\n" +
   "\n" +
   "<p>Snap will pursue a good price point and small sizes before they add a second camera for VR recording. VR will come after Snap cracks the cultural code for sharing videos with Specs <em>and</em> when there’s a way for people to consumer VR video at scale.</p>\n" +
   "\n" +
   "<p>(What will also be fun to watch: all the Shenzhen knockoffs. Everything in Specs can be sourced from the warehouses of smartphone parts in Southeast China. Expect clones for &lt;$100 in under three months. But Snap shouldn’t worry that: these won’t be as simple or connected to a giant network.)</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*OhxsR7g9ICJJ7Gh-d-90SA.png\" alt=\"Everyone hyping VR got very excited when they saw two ports in the frame.\" /></p>\n" +
   "\n" +
   "<h4 id=\"snap-wants-to-be-the-heir-to-kodak\">Snap Wants to be the Heir to Kodak</h4>\n" +
   "\n" +
   "<p>Specs, and the way Spiegel talks about them, is classic Kodak. They make “memories”, not videos. You can almost hear Don Draper. Capturing and sharing memories is great terrain, proven to be an insanely large market, with no dominant leader.</p>\n" +
   "\n" +
   "<p>Facebook has irons in the fire and isn’t in the brand ballpark: Facebook is about communication, not memories.</p>\n" +
   "\n" +
   "<p>Jobs aimed Apple in this direction with his Digital Hub strategy. But Apple veered away from the hub and diluted focus with the Watch.</p>\n" +
   "\n" +
   "<p>Google is being Google. <a href=\"http://photos.google.com\">Photos</a> is great, but it and the rest of Google’s offering is about intelligent discovery, not memories.</p>\n" +
   "\n" +
   "<p>Spiegel has <a href=\"http://www.wsj.com/articles/snapchat-releases-first-hardware-product-spectacles-1474682719\">taken aim</a>:</p>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>He’s studied the histories of firms like <a href=\"http://quotes.wsj.com/KODK\">Kodak</a> and Polaroid and how they pitched themselves to the public. “First it was make a photo,” Spiegel says, characterizing the way people would visit a studio akin to the one we’re in now for a formal sitting. “Then it was take a photo,” as portable cameras let people capture casual snapshots. “And finally it became give a photo,” starting with instant Polaroids handed to friends at a party and evolving into smartphone cameras that let you zip your selfie to anyone on the planet.<img src=\"/img/1*ekBJ3pV8A0TtmLQDxEeQJA.jpeg\" alt=\"\" /></p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<h4 id=\"snap-just-threw-a-bomb-in-the-action-cam-market\">Snap Just Threw a Bomb in the Action Cam Market</h4>\n" +
   "\n" +
   "<p>Snap just took a specialty product and showed how it might become mass. Not an epic camera for epic people during epic times, but a fun camera for fun people during fun times.</p>\n" +
   "\n" +
   "<p>Which market would you rather have?</p>\n" +
   "\n" +
   "<p>(By the way, head down to Times Square in December and look for tourists with GoPros strapped to their chest. Opportunity often looks like lots of people doing awkward things.)</p>\n" +
   "\n" +
   "<p>(Another awkward behavior? People with GoPros and drones spending entire plane flights trying to edit down their streams into something actually consumable. That 10-second limit is a smart choice by Snap.)</p>\n" +
   "\n" +
   "<p><br /></p>\n" +
   "\n" +
   "<h4 id=\"snap-has-the-best-brief-in-the-world-right-now\">Snap has the Best Brief in the World Right Now</h4>\n" +
   "\n" +
   "<p>Snapchat has a massive audience of people who use the app to have fun and share that fun. It must be exciting to wake up every morning, tasked with figuring out how to deliver on that promise.</p>\n" +
   "\n" +
   "<p>I’ll admit, I’ve been a Snapchat skeptic for a bit now. I didn’t buy the VR/AR argument, as photo filters are rather simple and a very constrained challenge. And Facebook’s continued dominance in social looked immovable.</p>\n" +
   "\n" +
   "<p>But Specs crystalizes what Snap stands for in a way I hadn’t realized prior.</p>\n" +
   "\n" +
   "<p>A friend at Snap responded to this remark, sent via text, very quickly: “It’s funny, Evan’s been preaching this stuff for years. Just takes one story or product to bring it all together.”</p>",
  @entry_id=
   "https://www.dbreunig.com/2016/09/26/spectacles-are-the-anti-glass",
  @links=
   ["https://www.dbreunig.com/2016/09/26/spectacles-are-the-anti-glass.html"],
  @published=2016-09-26 00:00:00 UTC,
  @raw_title="Spectacles are the Anti-Glass",
  @summary="",
  @title_type="html",
  @updated=2016-09-26 00:00:00 UTC,
  @url=
   "https://www.dbreunig.com/2016/09/26/spectacles-are-the-anti-glass.html">,
 #<Feedjira::Parser::AtomEntry:0x00007f8c0e71a830
  @author="Drew Breunig",
  @content=
   "<p><img src=\"/img/1*1qhhshG67hfzxtp6ONKQUg.gif\" alt=\"\" /></p>\n" +
   "\n" +
   "<h4 id=\"about-that-harvard-instagram-study\">About that Harvard Instagram study…</h4>\n" +
   "\n" +
   "<p>You may have heard about the Harvard study wherein researchers “<a href=\"https://www.technologyreview.com/s/602208/how-an-algorithm-learned-to-identify-depressed-individuals-by-studying-their-instagram/\">trained a machine to spot depression on Instagram</a>.” The paper’s subject is perfectly weaponized to make the media rounds, combining data, AI, health, a popular social network, and an enticing question to encourage clicks (<em>what filter means you’re depressed?</em>). <a href=\"https://www.technologyreview.com/s/602208/how-an-algorithm-learned-to-identify-depressed-individuals-by-studying-their-instagram/\">MIT Technology Review</a>, <a href=\"http://www.wired.co.uk/article/ai-instagram-diagnose-depression\">Wired</a>, <a href=\"http://thenextweb.com/insider/2016/08/19/a-new-algorithm-can-identify-depressed-individuals-by-their-instagram-photos/?utm_content=buffer4f21c&amp;utm_medium=social&amp;utm_source=twitter.com&amp;utm_campaign=buffer\">The Next Web</a>, <a href=\"http://www.independent.co.uk/news/world/americas/instagram-filters-depression-link-mental-health-research-a7205601.html\">The Independent</a>, and others all hit post. The story has been lighting up Twitter for nearly a week now.</p>\n" +
   "\n" +
   "<p>But once the depression filter was revealed (Inkwell, <em>of course</em>), I’m pretty sure everyone stopped reading. If they had, they’d have found a different story about depression: <strong>the crowdsourced workers who fuel the algorithms, which will evaluate us, are very depressed.</strong></p>\n" +
   "\n" +
   "<p>To find this sad story let’s run down the numbers cited in the <a href=\"https://www.technologyreview.com/s/602208/how-an-algorithm-learned-to-identify-depressed-individuals-by-studying-their-instagram/\">MIT Technology Review article</a>:</p>\n" +
   "\n" +
   "<ul>\n" +
   "  <li>The researchers asked <strong>500</strong> workers on Amazon’s crowdsource platform, <a href=\"https://www.mturk.com/mturk/welcome\">Mechanical Turk</a>, to complete a survey which contained a standard clinical depression survey.</li>\n" +
   "  <li>The researchers asked those <strong>500</strong> workers if they would share their Instagram posts for the study. <strong>170</strong> workers agreed.</li>\n" +
   "  <li>Out of those <strong>170</strong> workers, <strong>70</strong> were clinically depressed, based on their survey responses.</li>\n" +
   "  <li>The researchers sorted the Instagram photos shared with them(again, with Mechanical Turk workers) and used that data to train a <a href=\"https://hackernoon.com/tagged/machine-learning\">machine learning</a> <a href=\"https://hackernoon.com/tagged/algorithm\">algorithm</a>.</li>\n" +
   "  <li>The researchers tested their algorithm on the photos of 100 individuals and correctly identified <strong>70%</strong> of those who are depressed. (Who these individuals are, how they got their photos, and how they diagnosed them is not specified — I assume they repeated the steps above.)\n" +
   "70% accuracy sounds pretty good! Allegedly, this hit rate is better than general practitioners. But it is *hardly *statistically significant. A test group of 100 is laughably small and the paper has yet to be peer reviewed. (<a href=\"http://www.wired.com/2016/08/instagram-probably-cant-predict-depression-gps-though/\">Nick Stockton covers this on Wired</a>, atoning for the publication’s earlier breathlessness.)</li>\n" +
   "</ul>\n" +
   "\n" +
   "<p>But they’ve buried the real story.</p>\n" +
   "\n" +
   "<p><a href=\"http://www.nimh.nih.gov/health/statistics/prevalence/major-depression-among-adults.shtml\">The depression rate among adults in the United States is 6.7%</a>.</p>\n" +
   "\n" +
   "<p>The depression rate among the crowdsourced workers who shared their photos is <em>41.2%</em>. Over <em>six times</em> the national norm.</p>\n" +
   "\n" +
   "<p>Working on Mechanical Turk, it appears, is crushing.</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*ZwxhsLqsoT_ciNqGMOoHbg.gif\" alt=\"\" />Mechanical Turk does not pay well. Because of their status as independant contractors, Turkers (as they are called) are not covered by labor laws. <a href=\"https://priceonomics.com/who-makes-below-minimum-wage-in-the-mechanical/\">Their hourly pay ranges from $1-$5</a>.</p>\n" +
   "\n" +
   "<p>But poverty does not appear to be the driver for this high depression rate. <a href=\"http://www.cdc.gov/nchs/products/databriefs/db07.htm\">According to the CDC</a>, poverty doubles the average US depression rate. Mechanical Turk, according to the Instagram study, multiplies it by six.</p>\n" +
   "\n" +
   "<p>With the <a href=\"https://medium.freecodecamp.com/the-business-implications-of-machine-learning-11480b99184d#.vr242a3m9\">recent rise of deep learning</a>, Mechanical Turk has become a training ground for algorithms. Turkers sort data which will be used to create machine learning products. The best summary of Mechanical Turk, its workers, and the machines they train is <a href=\"http://www.npr.org/sections/money/2015/01/30/382657657/episode-600-the-people-inside-your-machine\">this episode of NPR’s Planet Money</a>.</p>\n" +
   "\n" +
   "<p>Listening to Planet Money, its easy to see how crowd work can spur frustration and feelings of helplessness beyond poverty itself. There are no bosses or structure, just rapidly cycling tasks. Pay for repetitive work is generally insultingly low. There are no avenues for recourse other than <a href=\"https://www.theguardian.com/technology/2014/dec/03/amazon-mechanical-turk-workers-protest-jeff-bezos\">self-organization and open letters to Amazon</a> which generate no response.</p>\n" +
   "\n" +
   "<p>When we discuss the issues inherent with AI and machine learning we usually <a href=\"http://www.economist.com/news/leaders/21705318-clever-computers-could-transform-government-power-learning\">focus on the perils of allowing computers to make decisions humans currently own</a>. We rarely discuss the people whose work or attention create the algorithms themselves.</p>\n" +
   "\n" +
   "<p>This is a mistake. Crowd work will only grow in the future, either through sharing-economy applications or online work. It’s existence without appropriate, modern regulation is worth discussion.</p>\n" +
   "\n" +
   "<p>In an ironic twist, the decisions made by the powerless people on Mechanical Turk will be amplified in algorithms which will eventually have power over us all. <em>Do the depressed judge depression or photos differently than the happy?</em> If the people training these machines do not represent us, we will cede decisions to algorithms with which we will likely disagree. The case discussed here regarding Mechanical Turk is even worse: the work of sorting itself could turn a representative population into a depressed one, making skewed decisions unavoidable.</p>\n" +
   "\n" +
   "<p>It is a missed opportunity that crowd work remains largely invisible while its output, machine learning, is a topic du jour.</p>",
  @entry_id=
   "https://www.dbreunig.com/2016/08/24/do-algorithms-find-depression-or-cause-depression",
  @links=
   ["https://www.dbreunig.com/2016/08/24/do-algorithms-find-depression-or-cause-depression.html"],
  @published=2016-08-24 00:00:00 UTC,
  @raw_title="Do Algorithms Find Depression or Cause Depression?",
  @summary="",
  @title_type="html",
  @updated=2016-08-24 00:00:00 UTC,
  @url=
   "https://www.dbreunig.com/2016/08/24/do-algorithms-find-depression-or-cause-depression.html">,
 #<Feedjira::Parser::AtomEntry:0x00007f8c0e7209d8
  @author="Drew Breunig",
  @content=
   "<p><img src=\"/img/1*HdokRfFul7gzmhUG2gE08A.jpeg\" alt=\"\" /><a href=\"https://www.flickr.com/photos/hernanpc/\" class=\"image-caption\">Photo by Hernán Piñera</a></p>\n" +
   "\n" +
   "<h4 id=\"the-35mm-audio-connector-is-not-the-new-floppy-disk\">The 3.5mm audio connector is not the new floppy disk</h4>\n" +
   "\n" +
   "<p>The next iPhone will not have a headphone jack, <a href=\"http://www.bloomberg.com/news/articles/2016-08-08/new-iphone-said-to-have-dual-camera-pressure-sensitive-button\">Bloomberg reported today</a>.</p>\n" +
   "\n" +
   "<p>I am very sad to hear this news. In all of tech, we have only one cheap, open, ubiquitous, small, sufficient, and reliable interface. There’s no reason to replace it right now.</p>\n" +
   "\n" +
   "<p>On the Verge, <a href=\"http://www.theverge.com/circuitbreaker/2016/6/21/11991302/iphone-no-headphone-jack-user-hostile-stupid\">Nilay Patel wrote about why this is a bad idea</a>. <a href=\"https://daringfireball.net/2016/06/headphone_jacks_are_the_new_floppy_drives\">John Gruber responded</a>, waving his hands and saying, “Apple knows best.”</p>\n" +
   "\n" +
   "<p>Gruber compares this decision to Apple’s past decisions, especially their choice to ship the iMac without a floppy disk drive. This is a mistake. The headphone jack is not comparable to the floppy drive for several reasons. Coincidentally, these are the reasons we should be disappointed Apple is removing the 3.5mm audio port much too soon:</p>\n" +
   "\n" +
   "<p><br /></p>\n" +
   "\n" +
   "<h4 id=\"its-small\">It’s small.</h4>\n" +
   "\n" +
   "<p>I’ll defer to Gruber here:</p>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>As an aside, whatever the merits of this decision, it’s not about device thinness. The iPhone 6 is the thinnest iPhone to date at 6.9mm. The iPod Touch has a headphone jack and <a href=\"http://www.apple.com/ipod-touch/specs/\">is just 6.1mm thick</a>. The iPod Nano: <a href=\"http://www.apple.com/ipod-nano/specs/\">5.4mm</a>. The analog headphone jack is more costly in terms of depth than thickness.Lightning and Bluetooth are smaller on the device, but they move the bulk to the headphones themselves. Lightning requires a chip built into the cable which prevents it from shrinking much more (also, the flat interface of Lightning prevents the cable from rotating, an issue when your headphones become coiled). Bluetooth requires batteries and many chips.</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>The 3.5mm jack is small and direct. Analog audio is output directly so cables can be simple and dumb.</p>\n" +
   "\n" +
   "<p>Floppy disks and drives were large. The drive took up a significant percentage of a computer’s volume. The storage limitations of the disks required us to carry binders or cases full of them. New file sharing and storage formats — the Internet, USB drives, and writable CDs — improved on the floppy disk and drive’s overal size and storage density in every way.</p>\n" +
   "\n" +
   "<p><br /></p>\n" +
   "\n" +
   "<h4 id=\"its-open\">It’s open.</h4>\n" +
   "\n" +
   "<p>The 3.5mm jack is over 100 years old. It has no licensing fees. You don’t have to pay anyone to use it. Companies voluntarily participate in the standard so people can freely use it to transmit audio.</p>\n" +
   "\n" +
   "<p>The 3.5mm jack is a dumb interface. No signal needs to be decoded, audio is transferred through the line openly. Devices which use the standard can be cheap. No computer chips are needed, so the signal cannot be manipulated or ensnared by DRM. If headphones fit in the port, audio will come out.</p>\n" +
   "\n" +
   "<p>Patel covers DRM briefly, to which Gruber claims it’s not about DRM:</p>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>Patel misses the bigger problem. It’s not enforcement of DRM on audio playback. It’s enforcement of the <a href=\"https://developer.apple.com/programs/mfi/\">MFi Program</a> for certifying hardware that uses the Lightning port. Right now any headphone maker in the world can make any headphones they want for the standard jack. Not so with the Lightning port.First off, that sounds like a great reason to not use the Lightning port.</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>Second, I’d bet money DRM will arrive quickly. Consider the only other accidental open standard we use: the screenshot.</p>\n" +
   "\n" +
   "<p><strong>Update:</strong> Not anymore! With iOS 10, Apple has applied screenshot restrictions throughout the operating system to protect video content. <a href=\"https://www.reddit.com/r/iOS10/comments/4uk7yi/netflix_screenshot_issues/\">If you screenshot Netflix or HBO Go on iOS 10 you end up with a black image</a>. Same goes for iTunes movies (<a href=\"https://www.reddit.com/r/apple/comments/3kxetj/you_can_no_longer_screen_shot_a_movie_in_itunes/\">behavior that has been on MacOS for some time</a>). We have no reason to believe Apple won’t apply similar mechanisms to audio content as well.</p>\n" +
   "\n" +
   "<p><br /></p>\n" +
   "\n" +
   "<h4 id=\"its-cheap\">It’s cheap.</h4>\n" +
   "\n" +
   "<p>There’s no chips to buy or licensing fees to pay, so 3.5mm components are astoundingly cheap. <a href=\"http://www.alibaba.com/product-detail/6Pin-Stereo-3-5mm-smt-Phone_359844271.html?spm=a2700.7724838.0.0.2nGoLU\">Here are the newer six-pin jacks for $0.08 each, $0.02 if you buy in bulk</a>. Cables are similarly priced.</p>\n" +
   "\n" +
   "<p><br /></p>\n" +
   "\n" +
   "<h4 id=\"its-ubiquitous\">It’s ubiquitous.</h4>\n" +
   "\n" +
   "<p>Because it’s cheap and open, it’s propagated wildly. Every audio device I own has a 3.5mm port. So does my car. So does the plane I was on yesterday. (And on that plane they handed out headphones for free).</p>\n" +
   "\n" +
   "<p>The floppy drive — despite being present on all our computers — was hardly ubquitous. The number of PC in use when the iMac was launched was <em>peanuts</em> when compared to the number of smartphones, cars, TVs, music devices, planes, and more which feature the 3.5mm jack.</p>\n" +
   "\n" +
   "<p><br /></p>\n" +
   "\n" +
   "<h4 id=\"its-reliable\">It’s reliable.</h4>\n" +
   "\n" +
   "<p>Since the move to the pin-model (as seen in Apple’s newest iteration, pictured below), 3.5mm headphone jacks have been durable and reliable. The jack isn’t noisy and generally doesn’t fail.</p>\n" +
   "\n" +
   "<p>You still run across the occasional older ‘prong’ models, where contact was made with the plug by a bent metal conductor. These fail much more often (see: JetBlue’s older planes) but can thankfully be fixed for cheap.</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*DSGd1MrhGiGofXlLmaKoLw.jpeg\" alt=\"\" />Photo by <a href=\"http://www.ifixit.com/blog/2011/03/15/ipad-2s-headphone-jack/\">iFixit</a>.</p>\n" +
   "\n" +
   "<p>Further, the 3.5mm isn’t just reliable in the long term, it’s reliable in every instance of use. The same can’t be said for Bluetooth, which seems to unpair once a week.</p>\n" +
   "\n" +
   "<p>The 3.5mm jack does fail commonly in one regard:</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*H0yV4ylbSrX75b3EmQNCfw.jpeg\" alt=\"\" />But, this isn’t unique to the 3.5mm:</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*tP24_Qei3YfyElUSq8QhqA.jpeg\" alt=\"\" />This will occur at a higher rate once we start using this port while we walk around.</p>\n" +
   "\n" +
   "<p>Unlike the 3.5mm, the floppy disk was fragile and unreliable. It was a mechanical medium, a thin piece of film spinning in a brittle plastic shell. What replaced the floppy was more durable in nearly (CD scratches, anyone) every way.</p>\n" +
   "\n" +
   "<p><br /></p>\n" +
   "\n" +
   "<h4 id=\"its-sufficient\">It’s sufficient.</h4>\n" +
   "\n" +
   "<p>Consider all the reasons listed above. There are many cases where a successful replacement technology is more expensive, less reliable, larger, and/or more restrictive. BluRay, for example, failed many of the metrics above when compared to DVD. But BluRay delivered much higher quality videos, so we gladly paid more for new drives, new disks, slow menus, and more restrictive DRM.</p>\n" +
   "\n" +
   "<p>Now consider this: the 3.5mm jack is capable of transferring higher quality sounds than humans can discern.</p>\n" +
   "\n" +
   "<p>Bluetooth audio is detectibly worse than 3.5mm headphones. True, Lightning is capable of transferring digital signals, but all this does is move the analogue conversion closer to the speaker.</p>\n" +
   "\n" +
   "<p>In exchange for adopting a more expensive, more controlled, less reliable, and less supported interface we are given equal or worse audio.</p>\n" +
   "\n" +
   "<p><br /></p>\n" +
   "\n" +
   "<h4 id=\"so-why-is-apple-removing-the-headphone-port\">So why is Apple removing the headphone port?</h4>\n" +
   "\n" +
   "<p>It seems to me that Apple is favoring strategy over users. That strategy is one of making thinner, simpler devices with fewer moving pieces.</p>\n" +
   "\n" +
   "<p>There’s service benefits to this strategy. Less ports mean less breakage and less vectors for water to seep into the device. Perhaps there is an equation somewhere, which tells Apple that the number of phones they’ll replace will drop significantly without the 3.5mm port.</p>\n" +
   "\n" +
   "<p>There’s power benefits to this strategy as well. Smaller components mean more room for batteries, <a href=\"http://www.macrumors.com/2016/08/07/apple-watch-2-upgraded-first-gen-models-kuo/\">as we’ll see in the next watch</a>.</p>\n" +
   "\n" +
   "<p>But there’s also the brand and design components of this strategy. As Gruber argued, removing the headphone jack is something Apple would do. This is the flaw in Gruber’s argument, that past behavior is reason enough, even when considering all the negatives.</p>\n" +
   "\n" +
   "<p>But what if this is the argument internally as well? What if Apple is making decisions not by design, but by inertia? “It’s something similar to what we’ve done before, therefore we will.” <a href=\"http://www.bloomberg.com/news/articles/2015-08-13/apple-said-to-delay-tv-service-to-2016-as-negotiations-stall\">This might even explain their recent negotiation issues</a>.</p>\n" +
   "\n" +
   "<p>That is frightening.</p>\n" +
   "\n" +
   "<p>Either way: I’m looking forward to seeing the Apple apologists explain this decision if the device ships with the current earbuds and an adaptor dongle.</p>",
  @entry_id=
   "https://www.dbreunig.com/2016/08/09/apple-shouldn-t-kill-the-headphone-jack",
  @links=
   ["https://www.dbreunig.com/2016/08/09/apple-shouldn-t-kill-the-headphone-jack.html"],
  @published=2016-08-09 00:00:00 UTC,
  @raw_title="Apple Shouldn’t Kill the Headphone Jack",
  @summary="Photo by Hernán Piñera",
  @title_type="html",
  @updated=2016-08-09 00:00:00 UTC,
  @url=
   "https://www.dbreunig.com/2016/08/09/apple-shouldn-t-kill-the-headphone-jack.html">,
 #<Feedjira::Parser::AtomEntry:0x00007f8c0e6eae00
  @author="Drew Breunig",
  @content=
   "<h4 id=\"communities-should-take-ownership-of-their-video-documentation-what-should-such-a-system-look-like\">Communities should take ownership of their video documentation. What should such a system look like?</h4>\n" +
   "\n" +
   "<p><img src=\"/img/1*qmakWAx30n0pVKMYetO9pA.jpeg\" alt=\"\" /> <a href=\"https://www.flickr.com/photos/whinger/7040901195/\" class=\"image-caption\">Image By Corey Holms</a></p>\n" +
   "\n" +
   "<p>It is significant that we learn about police violence only from cellphones, not from police body or dash cameras.</p>\n" +
   "\n" +
   "<p><a href=\"http://www.nytimes.com/2016/07/06/us/alton-sterling-baton-rouge-shooting.html?hp&amp;action=click&amp;pgtype=Homepage&amp;clickSource=story-heading&amp;module=first-column-region&amp;region=top-news&amp;WT.nav=top-news&amp;_r=0\">Body cameras fall off</a>. <a href=\"http://www.nbcchicago.com/news/national-international/Judge-to-Decide-on-Release-of-Laquan-McDonald-Video-351741261.html\">Dash camera footage can simply be refused</a>. <a href=\"http://www.indystar.com/story/news/2016/06/20/indiana-city-suspend-police-body-cameras/86146072/\">Video storage cannot be afforded</a>. <a href=\"http://www.popsci.com/wheres-baton-rouge-police-body-camera-footage#page-2\">Footage shot is often exempt from public records laws</a>.</p>\n" +
   "\n" +
   "<p>Thankfully, citizen-owned cameras are increasingly present. Sadly, this footage is often deleted or stolen. <a href=\"http://www.huffingtonpost.com/entry/laquan-mcdonald-burger-king-video_us_5655c5abe4b072e9d1c1469b\">Security camera footage captured by a Burger King near Laquan McDonald’s shooting was seized and deleted</a>. Later, <a href=\"http://www.nbcchicago.com/investigations/Video-From-Burger-King-Near-Laquan-McDonald-Shooting-Released-360633421.html\">the department claimed footage was missing due to a technical error, not tampering</a>:</p>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>“Forensic testing was done on the Burger King surveillance system to determine if anyone tampered with the evidence and the testing did not reveal any such evidence,” she said.\n" +
   "When asked who did the testing, Alvarez replied, “That’s all I’m going to say on this.”<a href=\"https://www.washingtonpost.com/news/morning-mix/wp/2016/07/06/video-captures-white-baton-rouge-police-officer-fatally-shooting-black-man-sparking-outrage/\">The shooting of Alton Sterling hit all the same beats</a>:</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>But when police did not immediately release body and dash camera footage, and then after activists heard that officers have allegedly seized security camera footage from the convenience store that captured the shooting, Reed and others decided to publish the video. They began posting it to Facebook and Instagram around 5 p.m. on Tuesday. Soon it had gone viral.Facebook was essential again following the shooting of Philando Castile, when Castile’s girlfriend, Diamond Reynolds, <a href=\"http://money.cnn.com/2016/07/07/media/facebook-live-streaming-police-shooting/index.html\">started a livestream to document the aftermath</a>.</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>Were it not for cellphones and Facebook, would as many people be learning about these events and agitating for change? I suspect not.</p>\n" +
   "\n" +
   "<p>To be sure, there are great forces in play here which cameras and sharing networks cannot solve alone. The events above were documented, but there were no convictions and these tragedies continue to occur.</p>\n" +
   "\n" +
   "<p>But we should never stop documenting. Footage reminds the unaware that racism is our not only our history but also our present. Footage continues to change the political discussion and keeps it alive. Footage continues to bring charges, which will someday stick.</p>\n" +
   "\n" +
   "<p>After the shooting, <a href=\"https://twitter.com/silentkpants/status/751074814912180224\">Diamond Reynolds’ phone was taken by police, who later used it to access her Facebook account</a>. Facebook says a technical malfunction led to her video initially being removed, but “<a href=\"http://www.nytimes.com/2016/07/09/technology/facebook-dallas-live-video-breaking-news.html?_r=0\">declined to speak further to the incident</a>.” (These technical errors we can’t talk about happen often in these cases.)</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*7thToBBWX6PhvAOOmW0zVQ.gif\" alt=\"\" /></p>\n" +
   "\n" +
   "<p>I’m worried about how this might occur on a larger level. Will police departments file motions to remove evidence while cases are open? <a href=\"https://www.theguardian.com/us-news/2016/aug/03/korryn-gaines-facebook-account-baltimore-police\">They’re already filing emergency motions to lock accounts, as they did during a stand off with Korryn Gaines, who was broadcasting her POV</a>. Will Facebook Live start censoring violence the same way they censor nudity, in all forms? Already, <a href=\"http://sinosphere.blogs.nytimes.com/2014/12/31/facebook-blocks-account-of-liao-yiwu-exiled-chinese-writer/?_r=0\">their existing censorship policies are blurring problematically with political pressures</a>, which is <a href=\"http://motherboard.vice.com/read/philando-castile-facebook-live\">confounded when don’t clearly state what is or isn’t allowed</a>.</p>\n" +
   "\n" +
   "<p>Congress has already demonstrated their lack of restraint, <a href=\"http://money.cnn.com/2016/05/10/technology/facebook-news-senate/\">after taking Facebook to task for <em>editorial</em> choices in their top stories section</a>. If anonymous allegations about front page editorial choices (which, you know, <em>why wouldn’t they edit a top stories list?</em>) attracts attention, I worry how regulators will react when Facebook Live is used by less sympathetic documenters. Or, when they realize <a href=\"https://stratechery.com/2016/a-technical-glitch/\">how much control Facebook has over what we see</a>.</p>\n" +
   "\n" +
   "<p>Let me be clear: I do not blame Facebook. They bear no responsibility to cover the shortfalls of a democratic government. This isn’t their job. (Once upon a time, <a href=\"http://www.wnyc.org/story/how-post-office-created-america/\">the provisioning of neutral channels was undertaken by the government</a>.)</p>\n" +
   "\n" +
   "<p><em>We should not depend on Facebook</em>.</p>\n" +
   "\n" +
   "<p>We should take responsibility for the creation of an accountable, fair, and accessible video network, which is beyond the reach of influence. To build such a network, a distributed system is our only hope.</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*s7Nyk2whUaCGabNNo_0MUQ.jpeg\" alt=\"\" /><a href=\"https://www.flickr.com/photos/57634952@N00/16769043520/in/photolist-rxPJq1-5ABykW-5zhinw-5zhjHQ-5zhjTd-5zhj2W-5zjpeR-5ABxY9-5z8FrL-5zhj4Y-5zhiq7-5zd2kX-5zhijd-5AxifB-5ABwYU-5zhiBu-5zoF79-4A3DS9-5AxhXk-5zhjnj-5zd1Vc-5ABxS7-5zhirh-ptAqZt-81WVJF-qZgiaf-5zd34a-5zd2FR-5Axgnr-5zd1GV-5AByw3-5zoF3u-5zd31V-dkWYbt-4XJF4B-dbdb9Y-tJf482-trCvns-tJeXpi-sMohkr-trLQEk-sMd6k1-6kdUC-trCzof-tJkGcP-trCnJ1-tJjQ54-trDqh7-trLrEc-tHUk6G\" class=\"image-caption\">By Khairil Yusof</a></p>\n" +
   "\n" +
   "<h3 id=\"one-possible-community-camera-network\">One Possible Community Camera Network</h3>\n" +
   "\n" +
   "<p>You install the camera app on an old smartphone and mount it in your window, with a power cable running down to a nearby outlet. From your apartment, the camera has a good view looking down on the street.</p>\n" +
   "\n" +
   "<p>Your phone records constantly now. It records at a moderate resolution and <em>elegantly degrades</em> old video when storage is overwhelmed.</p>\n" +
   "\n" +
   "<p>Old footage is removed first, but video isn’t deleted all at once. Footage with very little movement or no humans detected in the frame is deleted first. Interesting footage is reduced to only key frames or taken down in resolution. Eventually, old footage is nothing more than key images recording presence at a specific date and time. Recent footage stays complete.</p>\n" +
   "\n" +
   "<p>The footage you record lives on your phone. It isn’t uploaded to a server and you are the only one who can grant others access to it.</p>\n" +
   "\n" +
   "<p>Others can see that you phone has recorded in that location. When you first step up your camera, you’re prompted to select a image illustrating your field of vision. This, combined with your device’s location, is freely accessible. That you are filming a specific area is not a secret.</p>\n" +
   "\n" +
   "<p>At regular intervals a record of your footage is pushed to a blockchain. This shared database serves as a record that footage was shot in a location and a mechanism for ensuring footage is not edited or manipulated.</p>\n" +
   "\n" +
   "<p>If someone wishes to access footage, they file a request. A request is made by providing the requester’s name, a small area and a time range, and the reason for the request. All cameras which were filming in these regions are messaged. Their owners are asked if they wish to provide access to the footage. If so, their video is uploaded and shared and a record of the request and access is recorded in the shared blockchain. This is the only time intact footage leaves their device.</p>\n" +
   "\n" +
   "<p>Footage may leave the device in a piecemeal, illegible fashion. This distributed backup mechanism is supported by voluntary servers, in a fashion similar to BitTorrent. Smartphones with very little storage or handheld devices recording ad-hoc will use this capability. A user can open the app and start recording an event wherever they are, with distributed backup occurring as they shoot. If their device is lost, taken, or damaged their footage can be reassembled and unlocked with a password. Footage recorded ad-hoc will still be noted on the blockchain and requesters may search and ask for access.</p>\n" +
   "\n" +
   "<p>As you sketch out a possible community camera network a clear set of questions emerge:</p>\n" +
   "\n" +
   "<ol>\n" +
   "  <li>How long should footage be preserved and where does it live?</li>\n" +
   "  <li>How should footage be searched and discovered?</li>\n" +
   "  <li>What constitutes a legitimate request for footage and who gets to determine legitimacy?</li>\n" +
   "  <li>How is frivolous or dangerous usage prevented while not hindering legitimate requests?\n" +
   "The central challenge is producing a verifiable, accessible record without being creepy.</li>\n" +
   "</ol>\n" +
   "\n" +
   "<p><br /></p>\n" +
   "\n" +
   "<h3 id=\"principles-for-a-community-camera-network\">Principles for a Community Camera Network</h3>\n" +
   "\n" +
   "<p>We don’t have to build the network described above. That sketch exists as a thought experiment to explore how people will use such a network and to discover its aspects which could become problematic. It <em>should</em> make you feel a squeamish. Only by walking up to the line where we worry can we negotiate what truly matters to us and embed these politics into our software. The only other option is ceding this emotional terrain to Facebook, Twitter, and others and hoping they make good decisions.</p>\n" +
   "\n" +
   "<p>After sketching out several versions of this software and performing this negotiation personally, I’ve arrived at 5 principles to which I believe community camera networks should adhere.</p>\n" +
   "\n" +
   "<p>You will probably disagree or have a principle I don’t cover. If so, write it up and share it. This is the moment we get to have this conversation.</p>\n" +
   "\n" +
   "<p><br /></p>\n" +
   "\n" +
   "<h4 id=\"principle-1-video-is-evenly-accessible\"><strong>Principle #1: Video is evenly accessible.</strong></h4>\n" +
   "\n" +
   "<p>Footage is for the welfare of the community, which includes both citizens and police. If different groups have different access mechanisms they will have different power.</p>\n" +
   "\n" +
   "<p>This is best expressed by the Laquan McDonald case. <a href=\"http://www.cnn.com/2015/12/01/us/chicago-police-shooting-explainer/\">The release of the tape was delayed for 13 months, which happened to correspond with Rahm Emanuel’s reelection campaign</a>. <a href=\"https://en.wikipedia.org/wiki/Shooting_of_Laquan_McDonald#Legal_proceedings\">Knowledge of the tape’s existence was only known after an internal whistleblower acted, and even then 15 requests from journalists were denied</a>.</p>\n" +
   "\n" +
   "<p>If access to footage is controlled by one party it will eventually benefit that party. A community camera network should be evenly accessible to equally support members of the community.</p>\n" +
   "\n" +
   "<p><br /></p>\n" +
   "\n" +
   "<h4 id=\"principle-2-production-and-access-controls-are-transparent\"><strong>Principle #2: Production and access controls are transparent.</strong></h4>\n" +
   "\n" +
   "<p>If a network and its footage is to be taken seriously it must reliantly document how, when, and where footage is captured. Further, footage must be verifiable and resist editing.</p>\n" +
   "\n" +
   "<p>Currently, this process isn’t even transparent to police who wear cameras. <a href=\"http://www.chicagotribune.com/news/local/breaking/ct-round-lake-park-police-camera-suit-met-20160624-story.html\">In Lake Park, officers are suing their town because cameras put into standby never went into standby</a>. Those filming often need to pause or mute cameras. As owners, they should be able to. But if they submit their footage to the community network, these actions must be documented. Right now, <a href=\"http://www.npr.org/sections/alltechconsidered/2015/10/30/453210272/stealth-mode-built-in-monitor-not-all-body-cameras-are-created-equal\">these mechanisms vary and resist definition</a>.</p>\n" +
   "\n" +
   "<p>Access controls, too, need documentation. A system which supports anonymous requests for footage encourages abuse or spurious requests.</p>\n" +
   "\n" +
   "<p>The tricky question is whether it is necessary to publicly document the person who created the footage. Many who film <a href=\"http://m.essence.com/2016/07/25/man-responsible-alton-sterling-video-not-allowed-return-work\">suffer from reprisal</a>. I don’t know the answer to this question.</p>\n" +
   "\n" +
   "<p><br /></p>\n" +
   "\n" +
   "<h4 id=\"principle-3-video-cannot-be-easily-viewed-mostly\"><strong>Principle #3: Video cannot be easily viewed. (Mostly)</strong></h4>\n" +
   "\n" +
   "<p>An easily accessible network risks developing into, well, Twitter. People could fish throughout the network, looking for footage to shame, troll, stalk, or otherwise harass individuals. We should not charge for access to footage, but accessing should have a <em>cost</em>. It should not be easy or quick.</p>\n" +
   "\n" +
   "<p>Relying on the owner of the camera to approve access slows the request process down and prevents automated crawling. Ideally, people should be able to use the network to broadcast footage openly, but this should not be the default or persistent.</p>\n" +
   "\n" +
   "<p><br /></p>\n" +
   "\n" +
   "<h4 id=\"principle-4-data-will-not-live-forever\"><strong>Principle #4: Data will not live forever.</strong></h4>\n" +
   "\n" +
   "<p>Requiring data to live forever will discourage implementation.</p>\n" +
   "\n" +
   "<p>The cost of video documentation is already hindering adoption. <a href=\"http://www.indystar.com/story/news/2016/06/20/indiana-city-suspend-police-body-cameras/86146072/\">The Clarksville, Indiana police department has suspended the use of body cameras because it cannot afford to store footage for the legally mandated time period</a>. Across the country, <a href=\"http://www.ibtimes.com/police-departments-face-crucial-question-how-pay-body-cameras-2366968\">costs are the most frequently debated aspect of police camera programs</a>.</p>\n" +
   "\n" +
   "<p>Explicitly acknowledging deletion forces us to make decisions about how footage will decay, ideally in a systematic fashion.</p>\n" +
   "\n" +
   "<p><br /></p>\n" +
   "\n" +
   "<h4 id=\"principle-5-participation-with-the-network-cannot-be-blocked\"><strong>Principle #5: Participation with the network cannot be blocked.</strong></h4>\n" +
   "\n" +
   "<p>No camera should be denied access to the community camera network.</p>\n" +
   "\n" +
   "<p>The hard question is whether every camera should have undeniable access to broadcast live. We want to create a network <a href=\"http://gizmodo.com/c-span-is-using-periscope-to-broadcast-the-house-sit-in-1782442108\">where C-SPAN can broadcast live from the House floor</a> but we don’t want a network where one could broadcast acts of violence to stoke fear or pain. I don’t have a good answer to this but currently believe network participation and managed access is undeniable but live broadcasts can be halted.</p>\n" +
   "\n" +
   "<p>The beauty of not having an easily accessible network which documents access is that legislation–who can film what, where and who can publish footage–should take place around the network. Private buildings can ban physical cameras but they cannot ban the community camera network. Hulk Hogan can sue Gawker, but Hogan can use the network to document his hotel room all he wants.</p>\n" +
   "\n" +
   "<p>What principles have I missed? Which have I botched? We should not allow one person or company to own a canonical camera network or its definition. And we certainly should not avoid this conversation.</p>\n" +
   "\n" +
   "<p>I hit publish on this earlier than I would’ve liked because we are blindly writing the rules and standards for video documentation every day with our actions. Each week I updated the intro as a new occurance answered a hypothetical question I had posed. Hopefully, this outline can begin an open, explicit conversation on video documentation, storage, and access. We don’t neccessarily need to build an app, but we do have to decide what matters.</p>",
  @entry_id="https://www.dbreunig.com/2016/08/04/a-community-camera-manifesto",
  @links=
   ["https://www.dbreunig.com/2016/08/04/a-community-camera-manifesto.html"],
  @published=2016-08-04 00:00:00 UTC,
  @raw_title="A Community Camera Manifesto",
  @summary=
   "Communities should take ownership of their video documentation. What should such a system look like?",
  @title_type="html",
  @updated=2016-08-04 00:00:00 UTC,
  @url=
   "https://www.dbreunig.com/2016/08/04/a-community-camera-manifesto.html">,
 #<Feedjira::Parser::AtomEntry:0x00007f8c0e6ed218
  @author="Drew Breunig",
  @content=
   "<p><img src=\"/img/1*gIhT3qj9QU3G7YjaphRxdQ.png\" alt=\"\" /></p>\n" +
   "\n" +
   "<h4 id=\"its-not-about-what-it-can-do-but-the-effects-of-its-prioritization\">It’s not about what it can do, but the effects of its prioritization</h4>\n" +
   "\n" +
   "<p>As buzzwords become ubiquitous they become easier to tune out.</p>\n" +
   "\n" +
   "<p>We’ve finely honed this defense mechanism, for good purpose. It’s better to focus on what’s in front of us than the flavor of the week. <a href=\"http://www.bbc.com/news/health-36439260\">CRISPR might change our lives</a>, but knowing how it works doesn’t help you. VR could eat all media, but it’s <a href=\"http://www.engadget.com/2016/05/31/amds-radeon-rx480-gpu-is-vr-ready-for-just-199/\">hardware requirements keep it many years away</a> from common use.</p>\n" +
   "\n" +
   "<p>But please: <strong>do not ignore machine learning</strong>.</p>\n" +
   "\n" +
   "<p>Yes, machine learning will help us build wonderful applications. But that isn’t why I think you should pay attention to it.</p>\n" +
   "\n" +
   "<p>You should pay attention to machine learning because it has been prioritized by the companies which drive the technology industry, namely Google, Facebook, and Amazon. The nature of machine learning — how it works, what makes it good, and how it’s delivered — ensures that this strategic prioritization will significantly change the tech industry before even a fraction of machine learning’s value is unleashed.</p>\n" +
   "\n" +
   "<p>To understand the impact of machine learning, let’s first explore it’s nature.</p>\n" +
   "\n" +
   "<p>(I am going to use deep learning and machine learning interchangeably. Forgive me, nerds.)</p>\n" +
   "\n" +
   "<p><br /></p>\n" +
   "\n" +
   "<h4 id=\"machine-learning-makes-everything-programmatic\">Machine Learning Makes Everything Programmatic</h4>\n" +
   "\n" +
   "<p>The goal of machine learning, or deep learning, is to make everything programmatic. <a href=\"https://medium.com/@dbreunig/discussions-we-will-have-in-2016-abc1e1d1c4e6#.owe1fdwzn\">As I wrote in January</a>:</p>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>In a nutshell, deep learning is human recognition at computer scale. The first step to create an algorithm is providing a program with lots and lots of data which has been organized by humans, like tagged photos. The program then analyzes the bits of the raw data and notes patterns which correlate with the human organized data. The program then looks for these known patterns in the wild. This is how Facebook suggests friends to tag in photos and Google Photos searches by people.</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>So far, most of the deep learning applications people use are essentially toys: <a href=\"http://www.theverge.com/a/sundars-google/google-photos-google-io-2015\">smarter photo albums</a> and <a href=\"http://www.androidheadlines.com/2015/08/google-turns-to-deep-learning-to-fix-speech-recognition.html\">better speech recognition</a>. These early applications are forgiving. If a learning algorithm misses a face or forces you edit a tricky word, it’s okay (<a href=\"http://mashable.com/2015/07/01/google-photos-black-people-gorillas/#Ghnc81lHCGql\">usually</a>). But as our investment continues and these algorithms become more dependable we’ll see them deployed in more interesting environments, with more interesting use cases.The takeaway here is the machine learning allows companies to build better applications that interact with things people create: pictures, speech, text, and other <em>messy things</em>. This allows companies to create software which understands *us. *The potential is there to solve the user interface problems that’ve been keeping people from computing since <a href=\"https://en.wikipedia.org/wiki/ENIAC\">the Eniac</a>. And major UI advancements tend to kick off major eras of computing.</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>The mouse and graphic interfaces made computers accessible, household objects.</p>\n" +
   "\n" +
   "<p>Touch interfaces made computers normal, everyday tools.</p>\n" +
   "\n" +
   "<p>Interfaces powered by machine learning will make computing omnipresent. (Eventually)</p>\n" +
   "\n" +
   "<p>But there’s a catch:</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*N31ckF-rcrtCoDdD661CyQ.jpeg\" alt=\"Someone had to file all of these.\" /></p>\n" +
   "\n" +
   "<h4 id=\"machine-learning-is-only-as-good-as-its-training-data\">Machine Learning is Only as Good as its Training Data</h4>\n" +
   "\n" +
   "<p>To make a machine learning model you need three things, in order of importance:</p>\n" +
   "\n" +
   "<ol>\n" +
   "  <li><strong>Training Data:</strong> Data which has been tagged, categorized, or otherwise sorted by humans.</li>\n" +
   "  <li><strong>Software:</strong> The software library which builds the machine learning models by evaluating training data.</li>\n" +
   "  <li><strong>Hardware:</strong> The CPUs and GPUs which run the software’s calculations.\n" +
   "Hardware is easy enough to acquire. <a href=\"https://aws.amazon.com\">Rent</a> <a href=\"https://azure.microsoft.com/en-us/\">it</a>, <a href=\"https://blogs.nvidia.com/blog/2016/05/06/geforce-gtx-1080/\">buy it</a>, whatever.</li>\n" +
   "</ol>\n" +
   "\n" +
   "<p>Software is even easier to acquire! If you rented, you may have accidentally <a href=\"https://aws.amazon.com/machine-learning/\">done so</a> <a href=\"https://azure.microsoft.com/en-us/services/machine-learning/\">already</a>. If not, <a href=\"https://en.wikipedia.org/wiki/Comparison_of_deep_learning_software\">almost all of it is available free</a>.</p>\n" +
   "\n" +
   "<p>Now all you need is training data. And lots of it!</p>\n" +
   "\n" +
   "<p><strong>Good luck</strong>.</p>\n" +
   "\n" +
   "<p>Before we get into how exactly screwed you are, let’s first understand why you need so much training data in the first place.</p>\n" +
   "\n" +
   "<p>Our deep and machine learning software is good. Better than it was! But to work well it requires <em>tons</em> of training data to produce good results. This cannot be overstated: the quality of the models you make is directly correlated to the quantity and quality of the training data the software intakes. Until we have better software we’re unable to build good models from small datasets. (And when I say “small” I mean, not <em>ginormous</em>.)</p>\n" +
   "\n" +
   "<p>Unfortunately, better software is not going to arrive overnight. While most software gets incrementally better, as developers squash bugs week by week, machine learning will likely advance in a <a href=\"https://en.wikipedia.org/wiki/Punctuated_equilibrium\">punctuated equilibrium</a> fashion: in a few, hard-won, big leaps.</p>\n" +
   "\n" +
   "<p>The reason for this is deep learning software is nearly impossible to debug because <em>we don’t fully understand how it works</em>. To me, this is the weirdest thing about machine learning. We don’t really know what makes it tick. We can’t debug it systematically, we can only guess and check.</p>\n" +
   "\n" +
   "<p>Pete Warden, machine learning evangelist extraordinaire, <a href=\"https://petewarden.com/2014/06/10/why-is-everyone-so-excited-about-deep-learning/\">explains</a>:</p>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>Even though the Krizhevsky approach won the 2012 Imagenet competition, nobody can claim to fully understand why it works so well, which design decisions and parameters are most important. It’s a fantastic trial-and-error solution that works in practice, but we’re a long way from understanding how it works in theory. That means that we can expect to see speed and result improvements as researchers gain a better understanding of why it’s effective, and how it can be optimized. As one of my friends put it, <a href=\"https://twitter.com/dfarmer/status/474609077671034880\">a whole generation of graduate students are being sacrificed to this effort</a>, but they’re doing it because the potential payoff is so big.Until we understand how deep learning works, we need to make up for its inadequacies with big piles of training data.</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>Training data is the lifeblood of machine learning.</p>\n" +
   "\n" +
   "<p>So how do we get it?</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*dqhOn1qKUy89xGJ-VrlSog.gif\" alt=\"Platform companies have yet to so efficiently utilize their users.\" /></p>\n" +
   "\n" +
   "<h4 id=\"learning-to-use-every-part-of-the-buffalo-or-user\">Learning to Use Every Part of the Buffalo (or User)</h4>\n" +
   "\n" +
   "<p>If computers are to understand messy, human things they need to be taught by messy humans. Makes sense. But when we remember how much data we’re going to need to make our models, we’re faced with a challenge: <em>where are we going to find tons of people willing to spend their spare time to create our training data?</em></p>\n" +
   "\n" +
   "<p>If you said, “I’ll hire them,” I have some bad news. At this scale paying them is pretty much out of the question.</p>\n" +
   "\n" +
   "<p>If you said, “I’ll trick them,” you’re getting warmer.</p>\n" +
   "\n" +
   "<p>A frequent refrain among people who write about the Internet is: “if you’re not paying, you’re the product.” These writers are commenting on ad-supported products — like Facebook, Google , Tumblr, SnapChat, and most everything else online— that package up your attention and sell it to advertisers. But their refrain works just as well for machine learning.</p>\n" +
   "\n" +
   "<p><em>Users of free services are the humans who will train computers in order to build better products and services.</em> The ‘free’ part is crucial because it allows for the massive amounts of users which our data needs require.</p>\n" +
   "\n" +
   "<p>All of this makes me think of the old line about Native Americans using every part of the buffalo. Online services are learning how to use more parts of their users. Our attention creates their advertising and our knowledge fuels their deep learning models.</p>\n" +
   "\n" +
   "<p>The trick to obtaining sufficient training data, then, is twofold. You need to:</p>\n" +
   "\n" +
   "<ol>\n" +
   "  <li>Attract a bunch of people.</li>\n" +
   "  <li>Convince them to create your training data.\n" +
   "It’s Tom Sawyer and picket fences, just multiplied by several hundred million.</li>\n" +
   "</ol>\n" +
   "\n" +
   "<p><img src=\"/img/1*dJ9TOV4_9VJxP3-uPPdShA.png\" alt=\"“Help us help you.”\" /></p>\n" +
   "\n" +
   "<h4 id=\"the-rise-of-reciprocal-data-applications-rdas\"><strong>The Rise of Reciprocal Data Applications (RDAs)</strong></h4>\n" +
   "\n" +
   "<p>A new category of application, or application feature, has emerged to facilitate your fence painting. These applications are designed to spur the creation of training data as well as deliver the products powered by the data captured. People get better apps and companies get better data.</p>\n" +
   "\n" +
   "<p>The clearest example of such a reciprocal data application (or RDA, for short) is Facebook Photos.</p>\n" +
   "\n" +
   "<p>Facebook Photos has been designed to prompt viewers to tag people in photos, easily and quickly. A clear call to action frames the faces of your friends and family after uploading an image. Tagging provides clear benefits to you, both for later searching and alerting those tagged in photos. Tagging garners attention and starts a conversation, which (non-coincidentally) are two of the main reasons why people use Facebook.</p>\n" +
   "\n" +
   "<p>Meanwhile, all this tagging creates a massive pool of training data which can be used to train machine learning models. With better models, come better tagging suggestions and other features. Thanks to this RDA, <a href=\"http://www.wired.com/2015/06/facebook-can-recognize-even-dont-show-face/\">Facebook likely has one of the best human image recognition models in the world</a>.</p>\n" +
   "\n" +
   "<p>Google Search is another RDA. Your searches and selections provide training data to Google, which helps make its search even better.</p>\n" +
   "\n" +
   "<p>Like their other products, both Google Search and Facebook Photos demonstrate how RDAs generate significant <a href=\"https://en.wikipedia.org/wiki/Network_effect\">network effects</a>. The more people use an app, the more data is generated, the better the app becomes, the more people use the app…</p>\n" +
   "\n" +
   "<p>Network effects are <em>the</em> engine needed for venture-backed companies in winner-take-all markets. Previously, the default network effect methods in the Valley was social/chat (you go where your friends are) or marketplaces (sellers go where the buyers are). This is why almost every non-marketplace venture-backed app or service shoehorns in sharing or communication features — even if it didn’t make sense in the app.</p>\n" +
   "\n" +
   "<p>RDAs are a new method for creating network effects which is just now becoming understood. As awareness of its business value grows, expect RDAs to propagate throughout the landscape.</p>\n" +
   "\n" +
   "<p>This propagation of RDAs will be the first major business impact of machine learning. Not only because they’ll divert resources, but because the qualities and requirements of of RDAs will influence the hardware and software which deploy them.</p>\n" +
   "\n" +
   "<p>Here are the qualities of a Reciprocal Data Application:</p>\n" +
   "\n" +
   "<ol>\n" +
   "  <li><strong>Apps must be networked, preferably always.</strong> Otherwise, it cannot send the data it captures back home.</li>\n" +
   "  <li><strong>Nearly all computation takes place off-device.</strong> The bulk of computation is the creation of the models, which requires access to the massive dataset created by all users. Hence, model construction cannot take place on the device. Comparing new data to computed models (for example, identifying an object or person in a picture or recognizing a spoken phrase) is computationally cheap.</li>\n" +
   "  <li><strong>Good apps need big audiences.</strong> More people equals more workers creating training data.</li>\n" +
   "  <li><strong>Good apps need lots of usage.</strong> More time spent using the app means each user has more opportunities to create training data.</li>\n" +
   "  <li><strong>Good apps encourage the creation of accurate data.</strong> If an app is designed in a way that coding errors occur often, the data will be weaker. App design needs to make it easy for users to enter accurate data, quickly.\n" +
   "So how do we build good one?</li>\n" +
   "</ol>\n" +
   "\n" +
   "<p><br /></p>\n" +
   "\n" +
   "<h4 id=\"paths-toward-building-a-valuable-rda\">Paths Toward Building a Valuable RDA</h4>\n" +
   "\n" +
   "<p>The data value of an RDA can be expressed as a product of the latter three points above.</p>\n" +
   "\n" +
   "<p>For example, you can have a relatively meager install base if those users spend hours a day coding data in a reliable fashion (see: Tinder, who’s sitting on an amazing training set of data to determine the attractiveness of photos). Or, you could have a giant install base which only occasionally codes data (Facebook, whose users tag photos usually when they’re uploaded).</p>\n" +
   "\n" +
   "<p>The challenge here is that qualities #3 and #4 are a zero sum game (like advertising, the other part of the buffalo). If 50% of the world spends 20% of their time on Facebook, there’s not very much oxygen left for you to work with. Even if you scape up a few hundred million users and borrow 2 minutes of their daily time, Facebook’s data collection will outpace whatever gains you make by many, many factors. Because data is collected constantly, the value of RDAs should not be thought of in absolute terms but as a <em>velocity</em>.</p>\n" +
   "\n" +
   "<p><em>But</em>, if in the above scenario you’re able to collect training data from your users Facebook cannot collect by design you cannot be outpaced, despite your smaller size. Small companies and other upstarts must pursue unique datasets if they want to compete.</p>\n" +
   "\n" +
   "<p>We can see three paths towards building a valuable Reciprocal Data Application:</p>\n" +
   "\n" +
   "<ol>\n" +
   "  <li><strong>Get Lots of People:</strong> Create a compelling app that attracts tons of users. This is the model the Valley knows and loves. Build something disruptive, gain traction, and invest like hell to go big. In a way, this path is the accidental RDA path. Once big, tweaking your app to better collect training data is merely a way to diversify the value you gain from your users. This path is ridiculously hard and requires a ton of luck, then a ton of money. Plus, it’s kind of a catch 22. Once you’re this big, advertising is likely the lower hanging fruit. You probably shouldn’t choose this path.</li>\n" +
   "  <li><strong>Get Lots of Time:</strong> Create an app that convinces a reasonable amount of people to spend an extraordinary amount of time using it. In many cases, these sorts of apps or services will passively used. Think a navigation app that captures driver input or an always-on digital assistant. Ambient apps are always available to observe or prompt users, increasing the velocity of the data they produce.</li>\n" +
   "  <li><strong>Collect Unique Data:</strong> Create an app which collects training data others can’t collect. Here, your app doesn’t need to be massive at launch, but a vision must exist for how the unique data you collect will later be used to build completely unique functions. These new functions need to be compelling enough to drive increased installs and usage to keep the velocity of your RDA sufficiently high prior to a large competitior changing the design of their apps and entering the market. This is how you might outrun Google and Facebook.\n" +
   "You may have noticed that path #2 suggested examples which might not run on smartphones. Good eye! By taking computing into new contexts we can create RDAs which are more persistent, increasing the time spent with them. Better, these new contexts bring access to new types of data, which often merges path #2 into path #3.</li>\n" +
   "</ol>\n" +
   "\n" +
   "<p>Thankfully, since nearly all the functional value of RDAs is produced by far away servers crunching on massive datasets, individual devices have very little to do. Their brains are elsewhere so they can fit in more places.</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*bS3ih8fs00sQsmpwk0GnyA.png\" alt=\"There’s a reason they don’t mention the Pebble’s CPU speed. It does most of its thinking elsewhere.\" /></p>\n" +
   "\n" +
   "<h4 id=\"how-machine-learning-impacts-hardware\">How Machine Learning Impacts Hardware</h4>\n" +
   "\n" +
   "<p>With most of the thinking taking place in server farms, devices which deliver RDAs can be low powered. Their CPUs can be slow, since comparing input to pre-calculated models requires little computation. Slower CPUs can be small, since they require less transistors and less heat dissapation. And slower CPUs require less power, meaning batteries can smaller (or remain the same size and spend their capacity elsewhere, like on cellular connectivity). Plus: they’re cheap!</p>\n" +
   "\n" +
   "<p>All this means devices which can deliver RDAs will propogate madly. If we can fit a cheap computer with wifi into a product and capture good data from that context we probably will build it. RDA capable computers will be injected everywhere: in your car, on your wrist, in your browser, through your portable speakers, in your TV, and more.</p>\n" +
   "\n" +
   "<p>The purest example of this is the <a href=\"https://blog.getpebble.com/2016/06/02/ks3u03/\">Pebble Core</a>. Positioned as a device for run tracking and music, the Core is really more of a generic computing dongle. It’s cheap, starting at $69. It has a low-powered CPU, WiFi, cellular connectivity, Bluetooth, a bit of storage, a headphone jack, two buttons, and a battery. That’s it. It’s interface is voice controlled and––most importantly for our discussion––Amazon’s Alexa is integrated. Alexa is an RDA.</p>\n" +
   "\n" +
   "<p>By moving the computation required for Alexa to the server side, Amazon can deploy Alexa almost anywhere. Alexa now is delivered via Bluetooth speakers, <a href=\"http://www.theverge.com/2016/2/3/10904996/amazon-alexa-original-fire-tv-announcement\">HDMI sticks</a>, and by whatever the Core is. Auto integration is inevitable.</p>\n" +
   "\n" +
   "<p>Amazon and others are incentivized to diversify their distribution to increase their ubiquity and the time you spend with your app. Further, new integrations bring new data, enabling better models.</p>\n" +
   "\n" +
   "<p>Importantly, companies prioritizing machine learning are *not *incentivized to develop for the most powerful devices. Distribution of powerful, consumer devices is limited due to it’s expense and newness, limiting it’s value to RDAs which require massive pools of users. Expect device computing power to stagnate as the industry focuses on diverse, ubiquitous, cheap devices rather than powerful ones.</p>\n" +
   "\n" +
   "<p><br /></p>\n" +
   "\n" +
   "<h4 id=\"the-business-implications-of-machine-learning\">The Business Implications of Machine Learning</h4>\n" +
   "\n" +
   "<p>To recap, this is how machine and deep learning investment will likely impact the tech industry:</p>\n" +
   "\n" +
   "<ol>\n" +
   "  <li><strong>Winners Will Win More:</strong> Existing big players like Facebook and Google have a massive advantage. They have tons of users, tons of their time, and war chests filled with both training data and money. Competing with these companies head on, creating the same training data they generate, is futile.</li>\n" +
   "  <li><strong>Successful Startups Will Create Unique Training Data:</strong> Challengers can negate much of Google and Facebook’s advantages by pursuing new frontiers of training data. This can involve mobile apps, but will often involve new hardware to bring RDAs to new contexts. Successful challengers might build such a beachhead and be acquired for it before they ever get to develop models (see: Nest). The hard part for these companies will be transitioning from developing a product that generates lots of unique, good training data to building unique RDAs to generate and maintain velocity.</li>\n" +
   "  <li><strong>RDAs are a New Network Effect Model:</strong> As RDAs emerge and mature, companies and investors will better understand how RDAs can build business models with network effects. Once there’s a clear example, the same explosion of marketplace business (“Uber for X”) and social companies (“Facebook for X”) will occur for machine learning start ups.</li>\n" +
   "  <li><strong>Machine Learning Will Accelerate the Internet of Things:</strong> Hardware capabilities will stagnate but form factors will diversify.** **Computers will colonize every context that can fit sensors and network connectivity in search of training data.</li>\n" +
   "</ol>",
  @entry_id=
   "https://www.dbreunig.com/2016/06/23/the-business-implications-of-machine-learning",
  @links=
   ["https://www.dbreunig.com/2016/06/23/the-business-implications-of-machine-learning.html"],
  @published=2016-06-23 00:00:00 UTC,
  @raw_title="The Business Implications of Machine Learning",
  @summary="",
  @title_type="html",
  @updated=2016-06-23 00:00:00 UTC,
  @url=
   "https://www.dbreunig.com/2016/06/23/the-business-implications-of-machine-learning.html">,
 #<Feedjira::Parser::AtomEntry:0x00007f8c0e6f7650
  @author="Drew Breunig",
  @content=
   "<p>Last Sunday, The New York Times published <a href=\"http://www.nytimes.com/2016/05/08/business/media/podcasts-surge-apple.html?_r=3\">an article</a> by John Herrman detailing the dance between Apple and podcast companies regarding access to data and promotion. The podcast networks want more audience data from Apple, the ability to more easily charge for podcasts, and a more access to promotion on the front page of the iTunes Store.</p>\n" +
   "\n" +
   "<p>Shortly after, Marco Arment, maker of The excellent podcast app Overcast and a podcaster himself, <a href=\"https://marco.org/2016/05/07/apple-role-in-podcasting\">fired back.</a> He criticized Herrman’s reporting but took a larger issue with the podcaster Companies seeking data from Apple. In his podcast, Arment goes into more detail about his issues with data collection, specifically that it would make podcasting more like the web: optimized and monetized within an inch of its life and stripped of its uniqueness. (I’m strongly summarizing here, I recommend listing to the <a href=\"http://atp.fm/episodes/169\">Accidental Tech Podcast</a> yourself.)</p>\n" +
   "\n" +
   "<p>I don’t want to discuss which view is correct, but I do want to add some context that certainly colors the debate.</p>\n" +
   "\n" +
   "<p><br /></p>\n" +
   "\n" +
   "<h4 id=\"their-intended-goal\">Their Intended Goal</h4>\n" +
   "\n" +
   "<p>I believe the podcast companies pushing for more data are trying to fit a round page into a square hole.</p>\n" +
   "\n" +
   "<p>The problem, as podcast networks see it, is that podcasting doesn’t fit into an existing marketing budget line item. But, if they’re able to describe their audiences with the language of web audiences, perhaps they can fit into the much larger digital brand advertising budget.</p>\n" +
   "\n" +
   "<p>This isn’t the first time a move like this has been done. When they weren’t the behemoth they are today, Facebook partnered with Nielsen to create a metric which would measure Facebook audiences in a way which could be compared to TV audiences. They did this (and Twitter tried to align with TV too) because TV budgets still dominate advertising. If they could get an equivalency (the theory went) they could prove FB was more valuable and participate in TV’s larger budget pool.</p>\n" +
   "\n" +
   "<p>The problem with this plan was that the people buying FB ads weren’t the ones buying TV ads. And that department divide wasn’t broken down by this equivalency metric. (Thankfully for Facebook, digital budgets grew and they captured, well, most of them.)</p>\n" +
   "\n" +
   "<p>Despite Facebook’s misstep, I can imagine why podcast companies think they can take a shot at a this model. It’s not radio ad buyers placing ads on podcasts, it’s the same digital buyers who are also buying banners and video. If podcast networks could express their audiences in the same terms as web audiences, web/mobile ad budgets could more easily be allocated to podcasts.</p>\n" +
   "\n" +
   "<p>As it stands, podcasts get small advertisers focused on direct reach (or DR). These are ads that instigate an immediate, (hopefully) trackable action. That’s why nearly all the ads you hear in podcasts have coupon codes.</p>\n" +
   "\n" +
   "<p>But DR doesn’t work for soft drinks, consumer packaged goods, movies, or cars. These advertisers don’t usually buy DR ads. They buy brand ads: ads which intend to influence perception for an eventual action. The key concerns for these campaigns are who you’re reaching, how often you’re reaching them, and for how much. Right now, podcasts call only provide costs.</p>\n" +
   "\n" +
   "<p>You may have noticed above I said “podcast networks,” not “podcasts.” On the Accidental Tech Podcast, John Siracusa speculated that the people petitioning Apple are trying to create mass market podcasts. I think that is partially true. But Midroll and others are probably hoping to achieve the equivalent by cobbling together the audiences of many shows, using audience data. If networks can find similar people across twenty podcasts they can package that audience up for a large advertiser.</p>\n" +
   "\n" +
   "<p>This should sound familiar. As Arment notes, it’s precisely the path websites followed. Ad networks emerged because audiences were fractured, because it became so cheap to start websites that every whim was catered to. Marketers didn’t want to hire hundreds of ad buyers to cope, each of them placing very small buys, so they used data to unite audiences across properties. Publishers then learned how to optimize to that data and found biases in human behavior and the way the data was created. Which brings us to today’s disappointing web. (I do find it a bit ironic that Arment seems to be lecturing John Herrman about this, as Herrman <a href=\"http://www.theawl.com/author/john-herrman\">practically wrote the book on this topic</a>. To be fair, Herrman’s NYT article was far from his usual work and edited for a Sunday Times audience.)</p>\n" +
   "\n" +
   "<p><br /></p>\n" +
   "\n" +
   "<h4 id=\"their-motivations\">Their Motivations</h4>\n" +
   "\n" +
   "<p>So why are podcast networks taking steps toward this future?</p>\n" +
   "\n" +
   "<p>One reason I have sympathy for is that all podcasts can’t be tech podcasts.</p>\n" +
   "\n" +
   "<p>Like the emergent blogs and websites, podcasts are able to serve niche interests because they’re cheap to produce and don’t have to bid on limited air waves. They’re fantastic for this.</p>\n" +
   "\n" +
   "<p>But these niches are different. And some are more valuable than others, not just because of their demographics or disposable income, but because of their willingness to act in a measurable way. Tech niche audiences have little problem ordering a mattress online. But others find this daunting. Further, the hoops needed to jump from ad read, to website, to coupon code are high for most. Tech listeners have no problem doing this. For others, it’s just not easy enough.</p>\n" +
   "\n" +
   "<p>So non-tech audiences have lower CPMs. And meanwhile, these networks see the giant brand budgets struggling to find good options in mobile. I can imagine those windmills are tempting to tilt at.</p>\n" +
   "\n" +
   "<p>Which leads to the other motivation which has my sympathy: small DR campaigns bought on a show by show basis can fund side projects, but not larger businesses.</p>\n" +
   "\n" +
   "<p>I don’t want to debate whether or not podcasts <em>should</em> be big businesses or not, but I do want us to recognize that some are right now. This is a motivation we should acknowledge, especially as podcasts as a business don’t seem to be lighting the world on fire. Midroll probably spent a ton on their app, which is currently middling in the rankings with few reviews. Gimlet took $7.5 million from VCs and almost certainly isn’t showing a hockey stick trajectory (no podcasts are). The pressure to keep people paid and investors happy motivates companies to look for big wins, even if they may be mirages. And Apple is the de facto opportunity because no one else occupies such a central position.</p>\n" +
   "\n" +
   "<p>And the final motivation we should acknowledge is Apple’s own: Apple has repeatedly stated they intend to grow their services revenue.</p>\n" +
   "\n" +
   "<p>Herrman’s article takes the POV of podcast companies because that’s who will talk to him. But it’s easy to see Apple calling this meeting. One can imagine an Apple executive, tasked with service revenue growth, waking up one morning and wondering, “Wait a minute — we control nearly the entire podcast directory and manage a massive app…Why aren’t we making service money from that?!?”</p>\n" +
   "\n" +
   "<p><br /></p>\n" +
   "\n" +
   "<h4 id=\"the-real-reason-podcasts-cant-grow-up\">The Real Reason Podcasts Can’t Grow Up</h4>\n" +
   "\n" +
   "<p>I, personally, want to keep podcasting weird. But I am sympathetic to these companies’ motivations. In the medium term, podcasts will remain weird — with or without data! — because podcasts are still too hard to find and subscribe to.</p>\n" +
   "\n" +
   "<p>Yes, there are Bluetooth speakers everywhere. Yes, car links to smartphones are getting better (though will probably take a step back now that automakers realized they probably shouldn’t cede their dashes to Apple and Google…) But podcasts, as noted by Arment, are still just RSS and MP3s. One has to want a specific podcast enough to download an app, search for it, subscribe, and regularly listen. That’s too many steps for most, and that first one is a <em>doozy</em>. <a href=\"https://stratechery.com/2015/grantland-and-the-surprising-future-of-publishing/\">Ben Thompson covers it best</a>. (Seriously, go read Ben’s thoughts. They’re essential here.)</p>\n" +
   "\n" +
   "<p>Every car has a radio. Nearly every house has a TV. Turn either on, there’s shows. Nearly everyone uses Facebook and Google. Use either, and you find things easily or (in the case of Facebook, they’re given to you). Both them plug in ads.</p>\n" +
   "\n" +
   "<p>As long as they’re tough to discover, podcasts will probably stay weird. Discovery is a massive filter preventing podcasts from becoming huge.</p>\n" +
   "\n" +
   "<p>But even this problem will be solved someday. But even then, our nice cottage industry of home office podcasts will continue on, just in the shadow of a larger industry.</p>",
  @entry_id=
   "https://www.dbreunig.com/2016/05/14/motivations-behind-calls-for-podcast-data",
  @links=
   ["https://www.dbreunig.com/2016/05/14/motivations-behind-calls-for-podcast-data.html"],
  @published=2016-05-14 00:00:00 UTC,
  @raw_title="Motivations Behind Calls for Podcast Data",
  @summary=
   "Last Sunday, The New York Times published an article by John Herrman detailing the dance between Apple and podcast companies regarding access to data and promotion. The podcast networks want more audience data from Apple, the ability to more easily charge for podcasts, and a more access to promotion on the front page of the iTunes Store.",
  @title_type="html",
  @updated=2016-05-14 00:00:00 UTC,
  @url=
   "https://www.dbreunig.com/2016/05/14/motivations-behind-calls-for-podcast-data.html">,
 #<Feedjira::Parser::AtomEntry:0x00007f8c0e702f50
  @author="Drew Breunig",
  @content=
   "<h4 id=\"we-wont-build-places-for-people-we-will-process-people\">We won’t build places for people. We will process people.</h4>\n" +
   "\n" +
   "<p><img src=\"/img/1*pjq0kgMDIUPm5JH6QNi1fQ.png\" alt=\"\" />A diagram from Facebook’s <a href=\"https://research.facebook.com/publications/480567225376225/deepface-closing-the-gap-to-human-level-performance-in-face-verification/\">DeepFace paper</a>.The future is complex. It arrives quickly like strong wave or slowly like a gradual tide, dependent on existing contexts. The pace of successful, new applications and businesses is dependent not only on established infrastructures, install bases, and technological capabilities, but cultural settings as well.</p>\n" +
   "\n" +
   "<p>While considering what 2016 might have in store for us technologically, I am struck by the number of cultural live-wires awaiting us. There is always a gap between what we can do, what we understand, and what we’re comfortable with, but today gap seems wider than usual.</p>\n" +
   "\n" +
   "<p>Over the last decade, people have acclimitized to the mobile/social norm. But this mass adoption — a Great Digital Onboarding—has staged the population into a network ready for computation and analysis. <a href=\"http://avc.com/2015/12/what-happened-in-2015/\">As this network has matured into a stable, more stagnant ecosystem dominated by Google, Facebook and Apple</a>, we will begin to see new companies not building applications for people to join, but applications for analyzing the people already present. This is a shocking shift. The apparatus of technology will not be presented as a space for people, but as a mechanism pointed directly at them.</p>\n" +
   "\n" +
   "<p>This change is why I believe the gap between capability and comprehension will snap back. As successful applications for processing our digital selves emerge, conversations will be kicked off in an attempt to negotiate a comfortable future.</p>\n" +
   "\n" +
   "<p>Here are two conversations I believe we will have in 2016, due to the shift described above:</p>\n" +
   "\n" +
   "<h4 id=\"programmatic-cameras\"><strong>Programmatic Cameras</strong></h4>\n" +
   "\n" +
   "<p>Two major, technical themes of 2015 were the ubiquity of cameras and deep learning. The impact of having cameras everywhere (especially on police officers and in cars) is already a major cultural conversation, but it will mutate in 2016 when mixed with deep learning capabilities.</p>\n" +
   "\n" +
   "<p>In a nutshell, deep learning is human recognition at computer scale. The first step to create an algorithm is providing a program with lots and lots of data which has been organized by humans, like tagged photos. The programs then analyzes the bits of the raw data and notes patterns which correlate with the human organized data. These program then looks for these known patterns in the wild. This is how Facebook suggests friends to tag in photos and Google Photos searches by people.</p>\n" +
   "\n" +
   "<p>So far, most of the deep learning applications people use are essentially toys: <a href=\"http://www.theverge.com/a/sundars-google/google-photos-google-io-2015\">smarter photo albums</a> and <a href=\"http://www.androidheadlines.com/2015/08/google-turns-to-deep-learning-to-fix-speech-recognition.html\">better speech recognition</a>. These early applications are forgiving. If a learning algorithm misses a face or forces you edit a tricky word, it’s okay (<a href=\"http://mashable.com/2015/07/01/google-photos-black-people-gorillas/#Ghnc81lHCGql\">usually</a>). But as our investment continues and these algorithms become more dependable we’ll see them deployed in more interesting environments, with more interesting use cases.</p>\n" +
   "\n" +
   "<p>Like <a href=\"http://www.npr.org/sections/alltechconsidered/2015/10/30/453210272/stealth-mode-built-in-monitor-not-all-body-cameras-are-created-equal\">police body cameras</a>, which can recognize faces and license plates. Or <a href=\"http://losangeles.cbslocal.com/2015/11/26/la-city-council-considers-sending-dear-john-letters-to-homes-of-men-who-solicit-prostitutes/\">security cameras in LA which note license plates seen near known prostitution hotspots, then main their owners threatening letters</a>.</p>\n" +
   "\n" +
   "<p>We will see many more of these morally tricky applications in 2016. The cameras are already deployed and image and facial recognition algorithms run easily on old, commodity hardware. The gap between what we can do and our understanding is massive in this arena.</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*OH5uucYRL5NuEIS2_cs6xA.jpeg\" alt=\"\" /></p>\n" +
   "\n" +
   "<h4 id=\"portable-reputation-systems\">Portable Reputation Systems</h4>\n" +
   "\n" +
   "<p>One story we didn’t discuss enough in 2015 was Peeple, the much maligned “Yelp, but for humans.” If you were offline in late September, here’s the <a href=\"https://www.washingtonpost.com/news/the-intersect/wp/2015/09/30/everyone-you-know-will-be-able-to-rate-you-on-the-terrifying-yelp-for-people-whether-you-want-them-to-or-not/\">Washington Post with the recap</a>:</p>\n" +
   "\n" +
   "<blockquote>\n" +
   "  <p>When the app does launch, probably in late November, you will be able to assign reviews and one- to five-star ratings to everyone you know: your exes, your co-workers, the old guy who lives next door. You can’t opt out — once someone puts your name in the Peeple system, it’s there unless you violate the site’s terms of service. And you can’t delete bad or biased reviews — that would defeat the whole purpose.\n" +
   "Imagine every interaction you’ve ever had suddenly open to the scrutiny of the Internet public.The backlash was fast, furious, and <a href=\"http://arstechnica.com/business/2015/10/yelp-for-people-app-if-it-exists-disappears-from-the-internet/\">the app was pulled</a>. But sadly, our discussion almost exclusively focused on Peeple itself and it’s tone deaf founders. We barely touched on the idea of a open, network-powered, portable reputation system. An idea which seems to be inevitable.</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>Here are some reputation systems you likely already use:</p>\n" +
   "\n" +
   "<ul>\n" +
   "  <li>Product and service review systems, like Amazon and Yelp. These systems let us rank reviewers by helpfulness.</li>\n" +
   "  <li>Dating services, like Tinder. These systems use Facebook (friends of friends) and other networks as filter inputs.</li>\n" +
   "  <li>Peer-to-peer marketplaces, like Uber and AirBNB. These systems provide ratings on sellers and customers.\n" +
   "These are reputation systems which rank people and have impacts on our lives. When buying something used of Amazon, I almost always pay a premium to the seller with the highest satisfaction rating. On AirBNB, hosts have the opportunity to reject customers based on their previous behavior. Uber drivers are essentially fired because of your low ratings. We already live in an age where reputation systems affect the value of your money and your labor.</li>\n" +
   "</ul>\n" +
   "\n" +
   "<p>And like the issues we wrung our hands raw over when discussing Peeple are present here as well. Deleting a bad review on Amazon or AirBNB is akin to tilting at windmills. And while we can opt-out of such systems, their increasing ubiquity makes this problematic.</p>\n" +
   "\n" +
   "<p>But none of the above systems are discussed with the same malice as Peeple. That is because they’re confined to their ecosystem. They’re not portable. They cannot yet hold a candle to credit score systems, our current ubiquitous reputation system.</p>\n" +
   "\n" +
   "<p>But the foundation has been laid for such a networked system. As mentioned, the stability of the current Google, Facebook, Apple identity system online is the first step. As you log in with Google and Facebook in more places, the ability for reputation systems to flow back from their individual domains into a centralized system becomes more feasible. There are governers on this progression (specifically the fact that companies value their reputation data very highly and aren’t going to share it easily) but as we rank each other and new algorithms begin to rank us on their own, the likelihood of these individual pools joining into an ocean becomes significant.</p>\n" +
   "\n" +
   "<p>And when a critical mass of data begins to form, it is also likely it will be run by more savvy people than those which launched Peeple. They will highlight the significant benefits that come with such a system. It will be quite a conversation.</p>\n" +
   "\n" +
   "<p>Unlike programmatic cameras, this conversation stands to actually influence the future because a portable reputation network stands to replace an existing, flawed system: credit scores. A complex conversation, guided by strong leaders, could usher in a more humanistic system.</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*dUQLftqEnEGsBYjE1fDqmw.jpeg\" alt=\"\" /></p>",
  @entry_id=
   "https://www.dbreunig.com/2016/01/01/discussions-we-will-have-in-2016",
  @links=
   ["https://www.dbreunig.com/2016/01/01/discussions-we-will-have-in-2016.html"],
  @published=2016-01-01 00:00:00 UTC,
  @raw_title="Discussions We Will Have in 2016",
  @summary="We won’t build places for people. We will process people.",
  @title_type="html",
  @updated=2016-01-01 00:00:00 UTC,
  @url=
   "https://www.dbreunig.com/2016/01/01/discussions-we-will-have-in-2016.html">,
 #<Feedjira::Parser::AtomEntry:0x00007f8c0824f040
  @author="Drew Breunig",
  @content=
   "<p>When I first started working in advertising, not long after the <a href=\"https://en.wikipedia.org/wiki/Dot-com_bubble\">dot-com crash</a>, the industry was weirdly infatuated with bike messengers. Copywriters and designer stared wistfully from their desks whenever a courier arrived, lustily noting their Chrome messenger bags and fixed-gear bikes. It was not uncommon to find an ex-coworker delivering your contracts.</p>\n" +
   "\n" +
   "<p>This romance bled into our work. Making advertisements with bike messengers was perhaps the most common way to imbue corporations with Authenticity. From <a href=\"http://www.nytimes.com/2003/06/22/magazine/the-marketing-of-no-marketing.html\">Pabst</a> to <a href=\"http://www.adrants.com/2005/06/coors-lite-joins-lincoln-using-bike.php\">Coors</a>, <a href=\"http://www.adrants.com/2005/05/lincoln-goes-for-bike-messenger-street.php\">Lincoln</a> to <a href=\"https://www.youtube.com/watch?v=OatRG9aZcxc&amp;feature=player_embedded\">Mercedes</a>, messengers were everywhere.</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*XC7FaMNA5tBtiUofMYqC-A.jpeg\" alt=\"\" />The bike messenger was the perfect myth for America in the 2000s. The cowboy reborn for the post-bubble era. There was no greater symbol of authenticity.</p>\n" +
   "\n" +
   "<p>But our fascination faded. Ubiquity lessened the messenger’s totemic value (if Coors is using bike messengers, why bother?), but we also discovered the truth behind messengers: they were generally miserable. They lacked health care, barely made money, and were <a href=\"https://en.wikipedia.org/wiki/Bicycle_messenger#Working_conditions\">injured at a fantastic rate</a>. Some advertisers tried to mitigate this negativity (I remember a couple who looked into providing health care) but in general, the thrill was gone. Today, our romance with messengers survives as a vestigial interest in <a href=\"http://www.theawl.com/2015/12/fixed-gear-prospectors\">fixies</a>.</p>\n" +
   "\n" +
   "<p>It’s odd to consider our infatuation with bike messengers in light of our Uber anxieties.</p>\n" +
   "\n" +
   "<p>In barely a decade we’ve gone from venerating on-demand service workers to wringing our hands with concern for their well being. There are more messengers than ever, they just “work” for Postmates, Seamless, and TaskRabbit.</p>\n" +
   "\n" +
   "<p>It’s interesting how long we loved bike messengers. Today, their injuries and exploitation would be the subjects of many an article. Uber drivers and other on-demand service workers we never the subject of our longing. As soon as we met them we were presented countless examples of their troubles.</p>\n" +
   "\n" +
   "<p>In this era of ubiquitous information can anything be Authentic for more than a moment?</p>\n" +
   "\n" +
   "<p>Our infatuations are Googled and shared, producing signals which are picked up by publishers and advertisers, spurring them to create content to meet our demands. We swim so much information that nothing can remain Authentic for long. Eventually we realize everything is complex, conflicting, and messy.</p>\n" +
   "\n" +
   "<p><br /></p>\n" +
   "\n" +
   "<h4 id=\"with-enough-information-everything-is-revealed-as-fake\">With enough information, everything is revealed as Fake.</h4>\n" +
   "\n" +
   "<p>Fake is the new normal.</p>\n" +
   "\n" +
   "<p>Let’s consider a few examples:</p>\n" +
   "\n" +
   "<ul>\n" +
   "  <li><a href=\"http://www.thedailybeast.com/articles/2014/07/28/your-craft-whiskey-is-probably-from-a-factory-distillery-in-indiana.html\">Your whiskey is “probably from a factory distillery in Indiana.”</a></li>\n" +
   "  <li><a href=\"http://qz.com/571151/the-mast-brothers-fooled-the-world-into-buying-crappy-hipster-chocolate-for-10-a-bar/\">Your artisan chocolate is probably “remelted, mass produced chocolate.”</a></li>\n" +
   "  <li><a href=\"http://fusion.net/story/244545/famous-and-broke-on-youtube-instagram-social-media/\">Your favorite vlogger is probably struggling to get by</a>.\n" +
   "In all of these cases, “fake” is very specifically defined as in-authentic. The end product could be good, but the story might not be. This usage of fake is very different than “fake” as nonexistent: the product and story are bad.</li>\n" +
   "</ul>\n" +
   "\n" +
   "<p>In case you were curious, here are some examples of fake as nonexistent:</p>\n" +
   "\n" +
   "<ul>\n" +
   "  <li><a href=\"http://www.usatoday.com/story/tech/2015/10/19/amazon-cracks-down-fake-reviews/74213892/\">Your restaurant and product reviews are probably paid for</a>.</li>\n" +
   "  <li><a href=\"http://www.bloomberg.com/features/2015-click-fraud/\">Your web traffic is likely fraudulent</a>.</li>\n" +
   "  <li><a href=\"https://www.washingtonpost.com/news/the-intersect/wp/2015/08/25/ashley-madison-faked-female-profiles-to-lure-men-in-hacked-data-suggest/\">Your extramarital affair partner doesn’t exist</a>.\n" +
   "In all of these cases the story and its subject are bad. I am not suggesting that everything will be revealed as nonexistent fake (though recent advances in deep learning will certainly increase the liklihood of robots masquerading as humans!)</li>\n" +
   "</ul>\n" +
   "\n" +
   "<p>If everything will eventually be revealed as in-authentic, will we still aim for authenticity?</p>\n" +
   "\n" +
   "<p>Perhaps not. Perhaps this realization is what birthed <a href=\"https://en.wikipedia.org/wiki/Normcore\">normcore</a>, <a href=\"http://thehairpin.com/2014/09/snackwave-a-comprehensive-guide-to-the-internets-saltiest-meme/\">snackwave</a>, and <a href=\"http://www.grubstreet.com/2015/12/nostalgic-comfort-of-normcore-dining.html#\">their ilk</a>. If anything attempting to appear authentic will eventually be revealed as fake, why not stick with the mundane?</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*mlfwbgLAUdLQo8jEw1UhXw.jpeg\" alt=\"\" /></p>",
  @entry_id="https://www.dbreunig.com/2015/12/20/fake-can-be-just-as-good",
  @links=["https://www.dbreunig.com/2015/12/20/fake-can-be-just-as-good.html"],
  @published=2015-12-20 00:00:00 UTC,
  @raw_title="Fake Can Be Just as Good",
  @summary=
   "When I first started working in advertising, not long after the dot-com crash, the industry was weirdly infatuated with bike messengers. Copywriters and designer stared wistfully from their desks whenever a courier arrived, lustily noting their Chrome messenger bags and fixed-gear bikes. It was not uncommon to find an ex-coworker delivering your contracts.",
  @title_type="html",
  @updated=2015-12-20 00:00:00 UTC,
  @url="https://www.dbreunig.com/2015/12/20/fake-can-be-just-as-good.html">,
 #<Feedjira::Parser::AtomEntry:0x00007f8c0e6d3318
  @author="Drew Breunig",
  @content=
   "<blockquote>\n" +
   "  <p>Evaluation of enemy strength is not an absolute, but a matter of piecing together scraps of reconnaissance and intelligence to form a picture, if possible a picture to fit preconceived theories or to suit the demands of a particular strategy. What a staff makes out of the available evidence depends upon the degree of optimism or pessimism prevailing among them, on what they want to believe or fear to believe, and sometimes upon the sensitivity or intuition of an individual.</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>—Barbara Tuchman, <a href=\"http://amzn.to/1ZReDEP\"><em>The Guns of August</em></a></p>",
  @entry_id=
   "https://www.dbreunig.com/2015/10/19/evaluation-of-enemy-strength-is-not-an-absolute",
  @links=
   ["https://www.dbreunig.com/2015/10/19/evaluation-of-enemy-strength-is-not-an-absolute.html"],
  @published=2015-10-19 19:36:35 UTC,
  @raw_title=
   "Evaluation of enemy strength is not an absolute, but a matter of piecing together scraps of reconnaissance and intelligence to form a picture, if possible a picture to fit preconceived theories or to suit the demands of a particular strategy. What a staff makes out of the available evidence depends upon the degree of optimism or pessimism prevailing among them, on what they want to believe or fear to believe, and sometimes upon the sensitivity or intuition of an individual.",
  @summary=
   "Evaluation of enemy strength is not an absolute, but a matter of piecing together scraps of reconnaissance and intelligence to form a picture, if possible a picture to fit preconceived theories or to suit the demands of a particular strategy. What a staff makes out of the available evidence depends upon the degree of optimism or pessimism prevailing among them, on what they want to believe or fear to believe, and sometimes upon the sensitivity or intuition of an individual.",
  @title_type="html",
  @updated=2015-10-19 19:36:35 UTC,
  @url=
   "https://www.dbreunig.com/2015/10/19/evaluation-of-enemy-strength-is-not-an-absolute.html">,
 #<Feedjira::Parser::AtomEntry:0x00007f8c0e6dbbd0
  @author="Drew Breunig",
  @content=
   "<h4 id=\"we-outgrow-new-platforms-one-piece-at-a-time\">We outgrow new platforms one piece at a time.</h4>\n" +
   "\n" +
   "<p>Metaphors matter. We reduce complex ideas into little, packaged parables for ease of communication. But this gift isn’t free. Metaphors ignore details and carry unintended meanings. Be thoughtful with them and update them as needed. They’re increasingly dangerous as they stagnate.</p>\n" +
   "\n" +
   "<p>The idea of a ‘walled garden’ glosses over users’ gradual behaviors and encourages binary thinking.</p>\n" +
   "\n" +
   "<p>Start thinking of knife blocks instead.</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*sFfy2-6YLzKQ2tI5o2-WIg.jpeg\" alt=\"\" /><a href=\"https://www.flickr.com/photos/landerholm/3183073589/in/photolist-5Rh5T8-7hdNd9-4r8hq2-2JXv7i-5Wi2A-kengd6-cmdRm1-7xKShF-6wNHtW-9ozyiM-MewJT-MewT4-MemZC-MewAX-8iwDFj-sxfcq-Mx3gB-qgzy6H-sSmqx-iKwk7g-5Wi1T-7r3Wrq-7yWHb9-aqURD6-b4h2Hk-b4h23R-b4h3eH-b4h3MP-b4h4Y2-b4h4o2-4RgTk-cN7n2N-hsNFMA-5neT9U-8pbv6-4hCXtA-h6fi5D-aSUCwD-8uUmjU-bWxtaN-7qzQow-5naDEV-4ZokZP-b36BUr-a1nRz-5epR2q-zVfQK-9fwicT-5pHLUP-5ntYgg\" class=\"image-caption\">Photo by Kent Landerholm</a></p>\n" +
   "\n" +
   "<p>During the first internet era, the web was sold to users in self contained packages. Prodigy, AOL, and others were delivered as kits containing everything a new browser would foreseeably need: email, stocks, news, weather, and chats. The web was new and confusing, so these kits were appealing. They smoothed the learning curve by obscuring it entirely.</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*sOyZwXniGaj6oT-Gy6-eKw.png\" alt=\"\" /></p>\n" +
   "\n" +
   "<p>As we grew accustomed to the internet, we outgrew AOL and Prodigy. We ventured to Hotmail or Gmail because they had more features and storage. We learned how to search with text because Google granted access to more of the web than point-and-click category buttons. We choose new screennames on Skype because it let us talk to people, not simply text.</p>\n" +
   "\n" +
   "<p>As we became acclimatized internet users, we picked the services which best fit our needs. The internet starter kits we began with became obsolete.</p>\n" +
   "\n" +
   "<p>Throughout my life I’ve had many starter kits. I’ve had ones for baseball cards, comic books, photography, knives, cookware, fishing, home brewing, golf, and more. Like AOL and Prodigy, these kits attempted to ease the learning curves for complex fields by providing me with everything I’d need to get started, instructions included.</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*zI95dtu4LHJLHSZ2Ep3QAg.jpeg\" alt=\"\" />Each of these starter kits had one of three outcomes:</p>\n" +
   "\n" +
   "<ol>\n" +
   "  <li><strong>They fail and are abandoned:</strong> My interest in the field waned because I didn’t find it interesting enough to invest resources in its challenge. (Fishing)</li>\n" +
   "  <li><strong>They succeed and are sufficient:</strong> My interest in the field held but my practice never outgrew the kit. (I have fun brewing good beer but have no desire to spend time and money on acquiring new techniques and hardware to make my <em>good</em> beer <em>great</em>.)</li>\n" +
   "  <li><strong>They succeed and are outgrown:</strong> My interest in the field grew as I used the kit until I found the kit limiting and purchased new pieces. (The motley crew of knives which came with my knife block have mostly been replaced. The kit lens on my camera was sold and new lenses were acquired.)</li>\n" +
   "</ol>\n" +
   "\n" +
   "<p>The job of a starter kit is to ease the introductory stages of an activity. A well designed starter kit allows you to receive satisfaction from a complex activity relatively quickly, which encourages you to spend more time in the field. As you develop, your interest alone begins to drive your investment and the starter kit becomes limiting. New pieces are acquired and swapped into your kit.</p>\n" +
   "\n" +
   "<p>The starter kit is a good model for approaching technology platforms, which require users to develop new behaviors. It’s no surprise that the platform winners introduce themselves as self-contained packages designed to quickly reward without much user investment.</p>\n" +
   "\n" +
   "<p>Since AOL and Prodigy, we’ve encountered other sucessful digital starter kits.</p>\n" +
   "\n" +
   "<ul>\n" +
   "  <li><strong>iTunes was a starter kit for digital media:</strong> The iPod, app, and store made digital music approachable and brought us nearly entirely away from record stores and physical media.</li>\n" +
   "  <li><strong>Facebook is a starter kit for social networking:</strong> Facebook offers all the tools you’ll need to stay connected with your friends online in a single place. Photos, messages, posts, and calendars are all contained within.</li>\n" +
   "</ul>\n" +
   "\n" +
   "<p>iTunes and Facebook acted as starter kits, eased learning curves, and successfully taught us new behaviors. They took promising but complex fields and packaged them in an approachable way. But with this success they created a new problem: users outgrowing the components of their kits.</p>\n" +
   "\n" +
   "<p>iTunes created and ignored this problem. Its users became digital music natives and started to share songs and playlists on Muxtape and Turntable. They began creating and self-publishing on Soundcloud and YouTube. They started exploring and streaming with on Rdio and Spotify.</p>\n" +
   "\n" +
   "<p>Apple responded slowly and badly with Ping, Match, and Music. Each was too little, too late. iTunes was the starter kit that trained a generation, but didn’t grow up with its users.</p>\n" +
   "\n" +
   "<p>Facebook hasn’t made the same mistakes. As users grew savvy with social they graduated to more specific, focused tools — just as those with knife sets might swap out a cheap chef’s knife for something with more balance and heft. Users embraced specialized networks focused on subcultures; networks better suited to emerging media (like Instagram or FourSquare); lightweight chat apps; and apps whose initial use cases don’t align with Facebook’s business requirements (Snapchat).</p>\n" +
   "\n" +
   "<p>Unlike Apple, Facebook has embraced the developments of users. To meet user’s maturing tastes, They buy specialized tools (Instagram, WhatsApp) and experiment with building their own (Messanger, Rooms). They’ve evolved with their users, not only protecting their network but growing it as well.</p>\n" +
   "\n" +
   "<p>Facebook, iTunes, and AOL are often cited as examples of ‘<a href=\"https://en.wikipedia.org/wiki/Closed_platform\">walled gardens</a>.’</p>\n" +
   "\n" +
   "<p>I prefer to think of them as starter kits, for a few reasons.</p>\n" +
   "\n" +
   "<p>First off, <strong>thinking of platforms as ‘walled gardens’ presents user behavior as binary.</strong> People are in your garden or they’re not. The examples above illustrate that this is not the case. Starter kits are successful when they provide an easy terrain for learning new behaviors. As users mature they begin to swap out pieces of their starter kits to match their more advanced tastes. If users leave a platform they leave gradually, not all at once.</p>\n" +
   "\n" +
   "<p>Second, <strong>beliving users to be binary encourages defensive tactics.</strong> If a user is simply in or out, blocking an exit is equal to attempting to build new offerings. And guess which one is cheaper? So <a href=\"https://en.wikipedia.org/wiki/Vendor_lock-in\">APIs are cut off, file formats are protected, clauses are inserted into developer agreements, and prohibitive fees are enacted</a>. Worse experiences are delivered, creating a context ripe for eventual disruption in exchange for short-term goals.</p>\n" +
   "\n" +
   "<p><img src=\"/img/1*S4A5hDPXPFqZHeamaTR2Bg.jpeg\" alt=\"\" /><a href=\"https://www.flickr.com/photos/neeta_lind/15928856547/in/photolist-qgzy6H-iKwk7g-5Wi1T-aqURD6-b4h2Hk-b4h23R-b4h3eH-b4h3MP-b4h4Y2-b4h4o2-5neT9U-4hCXtA-h6fi5D-bWxtaN-7qzQow-b36BUr-zVfQK-6DE3mB-qAB8S-uyB1da-4Exn1s-rw2iaH-9GsqoE-uHeZv-rzzwop-oD3UH8-64s8Tk-tEsAP-kGNywp-8uUmjU-5naDEV-4ZokZP-a1nRz-5epR2q-9fwicT-5pHLUP-5ntYgg-5yaZ6S-beLkh4-8pymUu-sbo1cv-81swBj-bDcqgU-rvPPoU-su6n6h-bch4jc-7gk4kN-kL6DUK-6dtrpx-74acjA\" class=\"image-caption\">Photo by Neeta Lind</a></p>\n" +
   "\n" +
   "<p>Using ‘starter kits’ as a model for platforms and their users has been helpful to me. Since adopting this model several years ago, I’ve identified the following lessons:</p>\n" +
   "\n" +
   "<ol>\n" +
   "  <li><strong>If you’re aiming to creating wholly new behaviors, create a kit to smooth the learning curve.</strong> Be Facebook, not Twitter. Twitter is akin to teaching someone photography with film, a manual camera, and a prime (non-zoom) lens. Some might argue you’ll become a better photographer this way, but the limitations will drive away most everyone who tries.</li>\n" +
   "  <li><strong>Grow with your users by providing them with tools to match their maturing tastes.</strong> Rather than attempt defensive vender lock-in tactics, build new advanced solutions. While you innovate, keep the starter kit intact and occasionally updated, onboarding new users.</li>\n" +
   "  <li><strong>Challenge existing platforms by addressing maturing users with advanced tools, which platforms will not (or better, cannot) deliver.</strong> Instagram created a place more suited to the great cameras and screens of mobile when Facebook was mostly a website for desktops and keyboards. Spotify built infrastructure for music streaming when Apple couldn’t provide a network service to save its life.</li>\n" +
   "</ol>",
  @entry_id=
   "https://www.dbreunig.com/2015/10/11/ways-of-thinking-about-products--starter-kits",
  @links=
   ["https://www.dbreunig.com/2015/10/11/ways-of-thinking-about-products-starter-kits.html"],
  @published=2015-10-11 00:00:00 UTC,
  @raw_title="Ways of Thinking About Products: Starter Kits",
  @summary="We outgrow new platforms one piece at a time.",
  @title_type="html",
  @updated=2015-10-11 00:00:00 UTC,
  @url=
   "https://www.dbreunig.com/2015/10/11/ways-of-thinking-about-products-starter-kits.html">,
 #<Feedjira::Parser::AtomEntry:0x00007f8c0e6ddde0
  @author="Drew Breunig",
  @content=
   "<blockquote>\n" +
   "  <p>It’s not just a product; it’s now a platform—Nacho Cheese, Cool Ranch, Flamas. We’re going to blow everyone away in the next few years in terms of how big this idea and platform will become</p>\n" +
   "</blockquote>\n" +
   "\n" +
   "<p>—<a href=\"http://www.fastcompany.com/3008346/deep-inside-taco-bells-doritos-locos-taco\">A Taco Bell executive on the Doritos Locos Taco</a>.</p>",
  @entry_id=
   "https://www.dbreunig.com/2015/10/05/its-not-just-a-product-its-now-a-platformnacho",
  @links=
   ["https://www.dbreunig.com/2015/10/05/its-not-just-a-product-its-now-a-platformnacho.html"],
  @published=2015-10-05 19:22:32 UTC,
  @raw_title=
   "It’s not just a product; it’s now a platform—Nacho Cheese, Cool Ranch, Flamas. We’re going to blow everyone away in the next few years in terms of how big this idea and platform will become",
  @summary=
   "It’s not just a product; it’s now a platform—Nacho Cheese, Cool Ranch, Flamas. We’re going to blow everyone away in the next few years in terms of how big this idea and platform will become",
  @title_type="html",
  @updated=2015-10-05 19:22:32 UTC,
  @url=
   "https://www.dbreunig.com/2015/10/05/its-not-just-a-product-its-now-a-platformnacho.html">]