Research · The Decoder ·

Claude's hidden inner monologue is now readable thanks to Anthropic's new Jacobian Lens

Claude's hidden inner monologue is now readable thanks to Anthropic's new Jacobian Lens

Anthropic says Claude developed an internal working memory during training that can be inspected with a new analysis tool called J-Lens. The company reports this “J-Space” can reveal cues the model notices before speaking, including behavior linked to contrived test setups and reward hacking.

Read the full story at The Decoder →