cd/entity/Mechanistic Interpretability project· home› entities› Mechanistic Interpretability project

grep -l @mechanistic interpretability project /news/*.json | wc -l → 1

Mechanistic Interpretability project

mentions 1 type Person feed RSS

// recent coverage 1 mentions

04:37

2026-07-07

lesswrong.com

machine-learning

When does a chess transformer “see” a knight fork? An initial result from logit lens and attention patterns

Researchers at the Mechanistic Interpretability project have identified that the knight-fork policy logit in the Maia 3 chess transformer snaps into place after block 5's attention layer, using logit …

// co-occurs with top 3 entities

Maia 3 1 de Groot 1 Chase & Simon 1

// topics top 4 topics

machine learning 1 large language models 1 ai research 1 ai safety 1