ICML Spotlight Paper: Hundred-Neuron Circuits Can Steer Model Behavior

An ICML Spotlight paper provides the first empirical evidence that language model circuits are sparse in the neuron basis, as sparse as SAEs, showing that circuits of roughly a hundred neurons can steer model behavior.

2026-10-25 ~ 2026-10-25 · 2 related posts

• MLP Neurons Match SAE Sparsity: ~100-Neuron Circuits Steer LLM Behavior — burny_tech · 2026-10-25

1 near-duplicate retellings: burny_tech