---
format: "aidr-story-markdown/v1"
id: "37e3373021627d49c22a49585c000e1ba4165a8496dcb6469c75841266fc6380"
canonical_url: "https://aidr.today/37e33730?lang=en"
title: "Nick Cammarata Finds AI Models Legible Reversing Prior Consensus"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-10-15T16:40:07.000Z"
category: "Research"
topics: ["llm"]
source_urls: ["https://marketbrief.now/ai/nick-cammarata-finds-ai-models-legible-reversing-prior-consensus-ab687d62","https://x.com/nickcammarata/status/2110757819427574056","https://x.com/tszzl/status/2110758311339729061","https://x.com/AndrewCurran_/status/2110765687891489058","https://huggingnews.com/ai/nick-cammarata-finds-ai-models-legible-reversing-prior-consensus-ab687d62","https://www.techmeme.com/261015/p48"]
summary: "Research led by Nick Cammarata at Goodfire found that the vast majority of neurons in trillion-parameter-scale models are legible, enabling direct tracing of circuits in systems like Kimi K3. Audits of Qwen3-32B showed roughly 90% of neuron poles serve a single main purpose, reversing earlier estimates from Cammarata's OpenAI work that only 2% of neurons were legible. The findings may transform AI alignment into a normal science and increase the likelihood of achieving mechanistic interpretability before superintelligence."
---

# Nick Cammarata Finds AI Models Legible Reversing Prior Consensus

> [Open the canonical story](<https://aidr.today/37e33730?lang=en>)

**Published:** 2026-10-15T16:40:07.000Z
**Category:** Research
**Topics:** llm

## Summary

Research led by Nick Cammarata at Goodfire found that the vast majority of neurons in trillion\-parameter\-scale models are legible, enabling direct tracing of circuits in systems like Kimi K3\. Audits of Qwen3\-32B showed roughly 90% of neuron poles serve a single main purpose, reversing earlier estimates from Cammarata's OpenAI work that only 2% of neurons were legible\. The findings may transform AI alignment into a normal science and increase the likelihood of achieving mechanistic interpretability before superintelligence\.

## Sources

- [Story source](<https://marketbrief.now/ai/nick-cammarata-finds-ai-models-legible-reversing-prior-consensus-ab687d62>)
- [Story source](<https://x.com/nickcammarata/status/2110757819427574056>)
- [Supporting source](<https://x.com/tszzl/status/2110758311339729061>)
- [Supporting source](<https://x.com/AndrewCurran_/status/2110765687891489058>)
- [Story source](<https://huggingnews.com/ai/nick-cammarata-finds-ai-models-legible-reversing-prior-consensus-ab687d62>)
- [Story source](<https://www.techmeme.com/261015/p48>)