Monad released

PROJECT HERE: Monad

What is MONAD about?

Monad or Manifold Oriented Neural Agent Daemon as I jokingly called this hobby project, forcing an LLM to act out, an invisible to it, systems plan. Well the LLM does do a lot of heavy lifting still but a lot of the traidional agentic and harness stuff has been abstracted away to see how far the test of using an LLM as simply the language module. Built on my own theses and the Bayesian and Free Energy Principles

Project goal

The aim of the project was to explore how to simulate memory, decision-making based on past and predicted information, and the degree of autonomy such a system can exhibit. Using a language model, a limited demo environment was created for the Monad system to conduct conversations with users.

Interesting test results

Shows good promise in improving local AI agent systems through the prediction mechanism, even though data is intenionally degraded before the LLM input. System shows coherence and output shows the system quasi inferring it’s design upon multiturn usage, without pretrained data. The system was never told explicitly about the Monad system, or LLM for that matter, but never mentioned Qwen as model architecture.