Goodfire's Neuron Communities: a handful of neurons in LLMs can handle arithmetic and formatting
fatihdin4en · x · 2026-11-18
Interpretability startup Goodfire introduced "Neuron Communities" research, discovering communities of neurons inside LLMs that work together. They're surprisingly small but powerful — just a few neurons cooperate to perform common behaviors like arithmetic, ALL CAPS output, or recalling a date. Work led by research fellow Igor Shilov.
More from Research
- AIR-O decodes visual attention from brain scans in real time, generating images noninvasively — jonxuxu · 2026-11-18
- JAMA study finds AI-generated discharge summaries match clinician-authored ones on harm potential — EricTopol · 2026-11-18
- Hiding Random Data Points Is a Weak Test of Reaction Mechanisms, Chemists Warn — bravo_abad · 2026-11-18
- LoRA dial can precisely fake low benchmark scores: up to 97% of faked errors land on one letter — rohanpaul_ai · 2026-11-18
- Study finds latest Claude models were trained on protein sequences at scale — LeoTZ03 · 2026-11-18
- Reviewer fatigue in the coding-agent era: "everything is possible, nothing is interesting" — rishabh16_ · 2026-11-18