Goodfire's Neuron Communities: tiny neuron groups in LLMs drive behaviors like arithmetic

burny_tech · x · 2026-11-18

Interpretability startup Goodfire introduced "Neuron Communities" research: LLMs contain communities of neurons that work together, and they're surprisingly small yet powerful — just a few neurons cooperating can perform common behaviors like arithmetic, ALL CAPS output, or recalling a date. Work led by research fellow Igor Shilov. Retweeter banburismus frames it as a step toward higher-level abstractions in interpretability, since neurons frequently form functional groups rather than acting alone.

Related event: Goodfire Discovers Small Neuron Communities Driving LLM Behaviors(3 posts)→

Original post →

More from Models

Models channel →