interpretability

Anthropic Found Claude's Internal Monologue — And It Changes Everything

Anthropic discovered a 'global workspace' in Claude where the model silently reasons. It's interpretability's biggest breakthrough yet.

4h ago