Over the past four months, I've delved deeply into experimenting with LLM models. My goal? To simplify the development of reasoning chains without the need for an auxiliary programming language. This idea was inspired by studies on executive functions detailed in "Exploring Central Executive" by Alan Baddeley, 1996.
Baddeley described a system that manages cognitive processes and working memory, enhancing the retrieval of information from long-term memory:
>>> The central executive is responsible for controlled processing in working memory, including but not limited to, directing attention, maintaining task goals, decision making, and memory retrieval.
My idea is that a prompt could act as this 'central executive,' orchestrating all other agent instructions without defaulting to Python, etc.
To visualize this, I've developed a prompt structure: https://ibb.co/BBh8rHB
In this setup, the information within the green blocks is dynamic, altering based on the active prompt.
Prompts can interconnect through a reference table, allowing us to build a network where each prompt is a node. This framework can be expanded to include dynamically generated prompts that integrate into and expand the existing network during experimentation.
Github: https://github.com/turing-machines/mentals-ai
Feel free to ask any questions
No comments yet.
