MLP Neurons Match SAE Sparsity: ~100-Neuron Circuits Steer LLM Behavior
burny_tech · x · 2026-10-25
The author shares the ICML Spotlight paper showing MLP neurons are empirically as sparse a feature basis as SAEs, with a gradient-based circuit-tracing pipeline that finds causally effective neurons — 10² neurons control behavior on subject-verb agreement, and latent reasoning steps can be steered on multi-hop tasks.
Related event: ICML Spotlight Paper: Hundred-Neuron Circuits Can Steer Model Behavior(2 posts)→
More from Research
- UIUC's FAS benchmark trains LLMs to predict future research directions, +10.6% FAS — shangbinfeng · 2026-10-25
- Git pushes quadruple to 320M per quarter as vulnerability counts lag — LLMs said to boost code security — max_paperclips · 2026-10-25
- Leo Breiman's 2001 classic "Statistical Modeling: The Two Cultures" still resonates — miniapeur · 2026-10-25
- PRISM merge algorithm combines two Minimax-H3 finetunes, weights open-sourced — theOliviaRossi · 2026-10-25
- 8 Papers from Stengel-Eskin's Group at EMNLP 2026 Span Pragmatics, Faithful CoT, Agents — EliasEskin · 2026-10-25
- Q-CTRL demos AI-powered quantum sensors for GPS-denied navigation at MIT EmTech — MJBiercuk · 2026-10-25