TOPIC
2 Stories Published
Interpretability
Open Questions
17m
Do LLMs Have World Models? What Othello-GPT Shows and What It Doesn’t
Othello-GPT showed a sequence model can learn and use a hidden board state. Later work found such internal maps are often incoherent.…
Research
8m
Mechanistic Interpretability Under Test: Do Circuit Explanations Tell the Whole Story?
A master's-thesis preprint scores circuit-based explanations in GPT-2 Small. They read well and are sufficient, but cover only about 22% of the…