An interview with Goodfire researcher Nick Cammarata, whose viral blog post claimed LLMs are easier to understand than thought, as other researchers push back (Will Douglas Heaven/MIT Technology Review)

Key sources

1. SOURCE@nickcammarata · Oct 15, 10:40 PM · “I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had” · x.com
2. SUPPORT@choblin29 · Oct 15, 11:28 PM · “In a new Goodfire essay, he reports preliminary audits of Qwen3-32B finding that ~90% of neuron poles have one main job, ~7% have two related jobs, and only 2–4% remain unexplained.” · x.com
3. SOURCE@goodfireai · Oct 20, 3:48 AM · “I think our job is to sort of try to think like the neural network.” · x.com
4. SUPPORT@tszzl · Oct 15, 10:42 PM · “when nick told me that he had some interp breakthroughs, he was uh, underselling it. i think this may be remembered as a major moment in scientific history.” · x.com
5. SUPPORT@_nathancalvin · Oct 16, 12:30 AM · “Labeling neurons is not understanding behavior. Even if 90% of neurons have clear labels, explaining or ruling out something like deception means understanding how millions of them work together.” · x.com
6. SUPPORT@hosseeb · Oct 16, 4:52 AM · “Takeaway: superposition was a hoax. Neurons are actually simple and encode simple concepts; we just didn't know it because previous tools sucked.” · x.com
7. SUPPORT@teortaxestex · Oct 16, 8:21 AM · “you can see why frontier bros will be excited it's a permission structure to race ahead” · x.com
8. SUPPORT@jachiam0 · Oct 16, 4:58 AM · “I am excited to see the insights here translate into tools with quantifiable success rates for helping us identify potential harms or steer neural networks away from risk.” · x.com