FULL STORY
Neural Networks Are Already Clean: Hypothesis Meets Follow-Up
Nick Cammarata's essay arguing neural networks are already clean sparked pushback from interpretability researchers, followed by Goodfire's companion study suggesting most neurons are readable.
2026-11-10 ~ 2026-11-11 · 2 episodes · 6 posts
Episode 1 · Goodfire claims neural networks are already clean; interpretability researchers push back (2026-11-10, 2 posts)
Goodfire researcher Nick Cammarata argued in a lengthy post that polysemanticity stems from fixable issues and neural networks are already clean, prompting interpretability researcher burnytech to rebut that polysemanticity has not disappeared, sparking debate in the community.
- Interpretability researcher pushes back: polysemanticity isn't gone, you need all views — burny_tech · 2026-11-10
- Goodfire essay: polysemanticity is fixable, networks are already clean — burny_tech · 2026-11-10
Episode 2 · Goodfire Study: Most Neurons Are Interpretable, Cause for Optimism (2026-11-11, 4 posts)
Goodfire AI published the first technical follow-up to Nick Cammarata's "Neural Networks are Already Clean," using an automated interpretability pipeline to explain tens of thousands of neurons. The findings show individual MLP neurons are more interpretable and functionally limited than expected, suggesting a more optimistic outlook for interpretability.
- Goodfire finds most neurons are legible up to trillion-parameter models, backing Cammarata's essay — CSProfKGD · 2026-11-11
- Goodfire releases technical report on LLM neurons: understand them by computational function, not polysemanticity — aryaman2020 · 2026-11-11
- Goodfire's Curt Tigges: individual LLM neurons are far more interpretable than thought — The Cognitive Revolution · 2026-11-11
- Goodfire releases deep neuron interpretability study with autonomous Neuron Scientist agent — teortaxesTex · 2026-11-11