Goodfire releases technical report on LLM neurons: understand them by computational function, not polysemanticity
aryaman2020 · x · 2026-11-11
GoodfireAI released the first technical companion to Nick Cammarata's essay "Neural Networks are Already Clean", based on studying thousands of neurons with their novel automated interpretability pipeline.
Key claim: neurons are far more legible than previously thought, and the team now thinks of neurons by their computational function rather than whether they are polysemantic. The accompanying diagram shows their visualization format: each dot is one firing, the vertical axis is activation strength, width is damage per firing, lanes are modes, close colors hint at a possible shared function, and gray marks firings that fit no mode.
Related event: Goodfire Study: Most Neurons Are Interpretable, Cause for Optimism(4 posts)→
More from Research
- Data Colada Flags Fraud in Fifth Francesca Gino Paper, No Forensics Needed — danielrock · 2026-11-11
- Fable 5.2 tops GDPx chip-design benchmark, but task authors see zero gains — ycombinator · 2026-11-11
- Fortnow asks Claude about complexity in the Aronson-Ambainis proof: Promise-BQP-complete — fortnow · 2026-11-11
- Backdooring AI agents by reprogramming fewer than 10 dormant neurons — no training needed — DimitrisPapail · 2026-11-11
- Duke and Princeton paper: 10 small parallel agent teams beat best-of-N and one giant team — rohanpaul_ai · 2026-11-11
- Deletion Tuning: Zeroing 0.05% of Weights Matches Full Fine-Tuning With Less Forgetting — Thrumpwart · 2026-11-11