Safety
112d agoNatural Language Autoencoders Explain LLM Activations
Two LLM modules, an activation verbalizer and reconstructor, now map internal activations to text descriptions. This unsupervised method uses reinforcement learning to reconstruct residual stream activations. Researchers applied the tool during a pre-deployment audit of Claude Opus 4.6.
The Signal
It provides a concrete mechanism for model auditing by translating opaque internals into human-readable interpretations.