# Research

Long-term agenda: understand how AI models actually work internally.
Not benchmark scores, not behavioral phenomenology — the computational
mechanisms.

## Contents

- `fundamental-questions.md` — Five open questions nobody can answer
  about how LLMs work. These are the compass.
- `anthropic-j-space.md` — Summary of Anthropic's July 2026 paper on
  the global workspace / J-space in Claude models. Full writeup
  published: https://jayrup.me/blog/jacobian-lens-frequency
- `q1-prediction-to-reasoning/` — Deep dive on Q1: why next-token
  prediction produces reasoning. Why we can reverse-engineer toy
  models but not large ones. Sub-questions.
- `prime-grokking/` — Can a minimal architecture grok the next-prime
  function? Clean test case for whether neural networks can discover
  algorithmic structure from I/O pairs when memorization is easier.

## Related projects

- `~/Projects/oversight-vector/` — Measuring behavioral shift vectors
  under oversight (may pivot given J-space findings)
- `~/uni/dis/` — Dissertation on oversight cues and LLM deception

## Status

Pre-experiment. Identifying the right questions before designing
experiments. The Anthropic J-space paper may fundamentally change
the approach to oversight-vector — the J-lens provides direct access
to internal representations, making surface behavioral measurement
potentially obsolete.