My GAIB fork and some ideas that can be contributed back eventually I hope + my thoughts on media output presentation + more

[groovybits]

I have done this advanced like input for my chatbot GAIB (https://github.com/groovybits/gaib/blob/main/README.md) fork off of this project, implemented this dual input text box concept today. I wanted to share it and the code to help show how it can be more flexible like this in the future. I am adding modes like chat mode vs. the anime like display mode I currently am using, also there is a basic output mode too. I am interested in the output being more media modal oriented or flexible with the speaking output etc. That is my primary purpose with my fork, and hope to add back the features in a more generally useful way. The speaking and other features relying on pay services like Google could be adapted to use free methods for example.

https://github.com/groovybits/gaib/blob/main/utils/makechain.ts#L43

I've got it where it controls history length and manages long-term running + can run in a continuous query mode of questions. I am wanting to have it create a question from the previous output (WIP) like an auto-gpt type behavior. Right now it creates episodes evolving somewhat but could be better.

Also I am using different LLMs like GPT 3 vs. 4 for the document context generation of title/question vs. creation of the answer or story subsequent step. That is something that seems to be nice to do, faster / cheaper yet not the part that needs GPT-4, yet not sure if that is completely "right".

My token handling really looks at the counting of the output tokens and streams them still, so that feature that many want is shown. It allows a few things like making sure things are working still. One issue is that in this model I have to collect the output anyways to parse it into sentences / get gender and speaker information to give them voices that fit and different ones per character etc. so there is a delay inherently in getting that Anime style viewing subtitles and images generated theme. Yet to stream it of course is good to use for viewing it in another drop down area and I added it as a way to see status with a view of the current token being processed and "Loading..."

It would be nice to find others who can help with the various aspects, adding plugins perhaps to parse the doc returned information and augment it :D So many ideas, adding chat capabilities... more feed input... (perhaps a feed with document store backing like this gives the feed a nice memory), remembering past conversations in pinecone for history.... Using chroma / db that isn't firestore/firebase and user auth etc, not using stripe plugin for firebase.... (I leaned heavily on the easier side here and less work on coding, used GPT to code a lot too so turbocharged my time/output and figured these things can be done later after PoC working).

I am grateful for this project, without it I could not have gotten a start at this project!

I stream examples at https://twitch.tv/groovyaibot and have it running in Google cloudrun for people to try here:

https://gaib.groovy.org

Oh and I have the capability added, not enabled in my service (beta), to have subscriptions of tokens per month (allocation / management / proper counting not perfect yet). So it monitors usage and you can share it etc.

So this also allows you to monetize the chatbot and opens sources a method of doing that which does seem to work, I just haven't enabled it for I am not ready to handle running a pay service (plus it really needs more work).

There's also a sharing feature preserving the stories/answers produced. It has a global feed page as https://gaib.groovy.org/board which is turns it into a social media sharing site in a sense of the produced stories. Replay etc. seem like features not hard to add now, I added this last week, GPT helps speed up these things so much :).

[groovybits]

#272 (comment) I am refactoring my fork to no longer require extra services like firebase etc in regular / default mode. It can stream and uses the most current 104 version of langchain. I am also going to add the chroma support as an option to use instead of Pinecone, yet leave Pinecone in it too.