Nous Research Releases Contrastive Neuron Attribution (CNA): Sparse MLP Circuit Steering Without SAE Training or Weight Modification
Instruction-tuned language models refuse harmful requests. But which part of the model is actually responsible — and how does that…
