Related Project - Help Request

[RobertBateman]

Intro

For 18ish months, I have been experimenting and researching the best way to create large-scale multi-stage projects using LLM AI. This is similar in output to Auto-GPT as it aims to produce work that cannot be completed in a single request (or through a chat interface).

While I have made significant progress towards this goal, I have now encountered medical issues that mean I can no longer work on the project. I'm searching for someone to continue this work for me. I am prepared to make all my work open source hoping that it may be useful to someone and will not have been in vain.

Before I get into the 'what' let me first address the 'why' and share with you some reasons I might be worth listening to:

Here are my LLM AI Credentials:

• I have an Msc. in Software Engineering and over 15 years of experience as a programmer in the video game industry. In my career, I have been responsible for a few teams and projects as well as designing and authoring many tools for in-house content production pipelines.
  https://www.linkedin.com/in/robert-bateman-4064523a/

• I have been researching LLM AI since before April 2023, as shown by my old post here:
  Multi-Agent Model #979 (reply in thread)

• I participated in OpenAIs Evals project and gained API access to GPT-4 in May 2023 as seen here:
  [Eval] An array of Liar Paradox-based evals openai/evals#883 (review)

• I made a serious attempt to create a company to capitalise on the new tech I was developing:
  https://www.aicosmosstudios.com/
  https://www.linkedin.com/company/ai-cosmos-studios-ltd/

• I had a multi-agent collaborative network operational in July 2023
  https://www.youtube.com/watch?v=Jf90db9_7aM

• The collaborative network eventually made a short story
  https://www.linkedin.com/posts/ai-cosmos-studios-ltd_our-multi-agent-collaborative-network-is-activity-7096050202800726016-0rEP

• I then changed direction from using agents to a bespoke hybrid format I invented. To keep my innovations secure I stopped posting or sharing my progress.

• Also my mind is very unusual: I have ADHD, Autism, Dyslexia and an IQ in the 99th percentile.

Project Summary

Thought Tree AI

The project I have been working on I have named 'Thought Tree AI' and also secured the web domain for it: https://www.thoughttree.ai/. Although I haven't done anything with it yet as I have found myself struggling to pen documentation, much less make a presentable website out of it.

Overview

Initially, the project aimed to fully automate content creation for media based on unspecified arbitrary requirements that would only be known at execution time. This would facilitate the business goal of providing a service to produce bespoke entertainment media on request.

Once I had designed a system that could do this, I realised that this was only a small fraction of its capabilities. By examining the multimedia production pipelines of the video game industry and boiling them down to the principle relationships, I created a language that could describe the underlying thought processes.

In essence, I have created a programming language for thought processes. Thought processes can now be modularised, isolated and iterated on just as we do for conventional programming code. The ability to serialise and version control a thought process also allows for collective sharing and improvement. I dream that such modules could be hosted on a site like https://huggingface.co/ one day.

Design Documents

I have worked hard to explain my design, but as my medical issues have worsened I have not been able to complete my documentation. This has proven to be a significant problem as the ideas underpinning the design are difficult to articulate, involving a complex mix of high-level conceptual ideals and low-level programming logic.

Here is a link to an export of the latest WIP documentation:
https://drive.google.com/file/d/1tvlm4nJPMzeaI36kNwxuq-v1BV-lMHQe/view?usp=sharing

It's not finished and lacks a lot of explanations for the reasons behind various choices made. This also means some benefits of the approach or design have not been sufficiently explained or highlighted and are effectively hidden.

What Help Is Needed

There are 3 main areas I would like to request help with:

Documentation

I am not a scientist or researcher I work in video games. My documentation reads more like a TDD, a user guide or something in between. I am hopeful someone would be willing to attempt to rewrite my documentation in a more conventional way allowing it to be shared and understood by a wider community.

If anyone feels like expanding on my explanations with better diagrams, wording or structure, let me know and I can add a link to your work on the website (https://www.thoughttree.ai/).

Cognitive Engines

In my documentation, I show that one of the ideas is that programs written to execute the thought tree modules (such as TTML) are referred to as 'Cognitive Engines'.
The Cognitive Engines use nothing special, only conventional program code. They will need to integrate with LLM providers' APIs, other than that, they should be reasonably straightforward to build.

I am not a master of all programming languages, and I have only been working on an example Cognitive Engine in c#. The example is unfinished, much like the documentation. I am happy to share the repo for it: https://github.com/AI-Cosmos-Studios-Ltd/PreProd/.
The original design philosophy allows many different kinds of Cognitive Engines to exist. So they could be tailored to specific needs or authored to a required standard without compromising the ability to read serialised thought tree modules like TTML.
Anyone can make a Cognitive Engine that accepts a TTML file. Which LLM provider it connects to and how that's managed may differ.

I hope that someone may be interested enough in the idea that they could build one to try the concept out.

Further Research

I feel my research is only the tip of the iceberg. There are a myriad of ways that this can be expanded upon or improved. There is even potential for a limited form of recursive self-improvement.

I want someone to try to do something interesting with this concept. It's too interesting to go to waste.

Conclusion

Thank you to those who have read this far, I am humbled you have taken the time to hear me out. I am not actively working on this project and my ability to respond may be slow or impaired.

Please feel free to use any and all of the information I have shared in any way you wish, without reservation or restraint. While attribution would be nice I will not require it, so this work is usable in the broadest scope possible.

If you can't help directly, please consider sharing to help reach more people.

All the best,

• Rob