When Self-Reference Fails to Close: Matrix-Level Dynamics in Large Language Models

Create an observation tool that monitors internal matrix dynamics of open-source models like Llama or Gemma to debug self-referential failure cases.

Suggested repo: matrix-vision

"See deep inside your LLM when it starts thinking about itself."

Estimated effort: 55h