Nous Research Releases Contrastive Neuron Attribution (CNA): Sparse MLP Circuit Steering Without SAE Training or Weight Modification
No SAE training. No weight modification. Nous Research just cracked sparse circuit steering.
Why it matters
Nous Research introduces CNA, a novel method for steering LLM behavior through neuron circuit identification and ablation without the computational overhead of sparse autoencoders or model weight changes. This reduces friction for interpretability-driven control and opens safer, lower-cost paths to model behavior modification.
The key facts
7 to knowMethod: Contrastive Neuron Attribution (CNA) for sparse MLP circuit steering
No sparse autoencoder (SAE) training required
No weight modification needed
No degradation to general capability benchmarks
Targets sparse MLP neuron circuits for LLM behavior steering
Published by Nous Research (via MarkTechPost)
Date: May 23, 2026
Go to the source
MarkTechPostmarktechpost.com
Publisher excerpt: Nous Research releases Contrastive Neuron Attribution (CNA), a method that identifies and ablates sparse MLP neuron circuits to steer LLM behavior — no sparse autoencoder training, no weight modification, and no degradation of general capability benchmarks. The post Nous Research Releases…