Goodfire finds neuron communities via gradient attribution, then ablates them to verify causality
burny_tech · x · 2026-11-18
Interpretability company Goodfire describes a method for discovering behavior-linked neuron communities inside models without specifying behaviors in advance.
Pipeline:
- Run gradient-based attribution over millions of prompts to identify influential neurons
- Connect neurons frequently attributed together into a graph, then run community detection
- Test each community for necessity and sufficiency, and probe which behaviors it supports
The approach enables large-scale, hypothesis-free discovery and causal verification of functional units inside models.
More from Research
- Voyager 1 Reaches One Light-Day From Earth, Signals Now Take 24 Hours — snowclipsed · 2026-11-18
- TCS Hiring Statement Urges Quality Over Quantity as AI Fuels Paper Arms Race — fortnow · 2026-11-18
- New paper: a sociocultural approach to linguistic creativity — abenitezburraco · 2026-11-18
- 27B Model Rewrites Its Own Weights Without Training or Gradients, Paper Claims — udmrzn · 2026-11-18
- From 'impossible' to a working Bluetooth beacon: open-source EDA chip design is real — IgorCarron · 2026-11-18
- AURA 2.0: a streaming video LLM that speaks first, calls tools, and grounds visually — Shuhan Li · 2026-11-18