Goodfire's Nick Cammarata claims neural networks are already clean, sparking interpretability buzz
TheZvi · x · 2026-10-19
- Nick Cammarata of Goodfire published "I Can See Everything," boldly claiming "Neural Networks are Already Clean" — that model internals are directly readable — followed by a follow-up post addressing the ensuing discussions.
- Accompanying visuals include a correlation matrix from Anthropic persona evals on Qwen3-8B showing which traits share neurons (virtue/honesty traits cluster; Machiavellianism/psychopathy anti-correlate), and a neuron-level decomposition of Kimi-K3 answering a found-wallet prompt: facts (lost property, honesty) push toward "return," with the strongest push surprisingly from neurons holding the answer back.
- Reactions are euphoric: OpenAI's Dean W. Ball called it "perhaps one of the most important single days in deep learning history," and optimists compare it to "Attention is All You Need," saying mechanistic interpretability may be solvable or nearly solved.
- TheZvi frames the key question: how big an update is this really, and what does it mean going forward?
Related event: Ex-OpenAI Researcher's Essay: Neural Networks Are Already Clean(5 posts)→
More from Research
- 1-gram wireless mesoscope images cortex-wide activity in freely moving mice — KordingLab · 2026-10-23
- Reversible computing veteran poses open problem: universal BARCS primitives from unbiased Josephson junctions — MikePFrank · 2026-10-23
- Arc Institute uses genomic language model to map 376M prokaryotic noncoding sequences into 5.14M clusters — tedgreenwald · 2026-10-23
- The rise of AI-assisted peer review: reviewers are making it into the appendix — abursuc · 2026-10-23
- Anthropic uses Claude to formally verify security properties in 100+ open-source libraries at RISC-V assembly level — AnthropicAI · 2026-10-23
- Researchers test 9 strategies to pick LLM teams, beating random selection by up to 36.1% — shangbinfeng · 2026-10-23