Joshua Batson

Batson studies how large language models represent concepts and make decisions internally. At Anthropic he helped scale sparse-autoencoder methods that expose interpretable features inside Claude, advancing efforts to understand and monitor otherwise opaque model behavior.