/TECHStory update pending

Anthropic Research Uncovers Global Workspace Mechanism Inside Claude

Source Posts on X

Cluster sources

I often find in Claude’s reasoning traces it will have thoughts / opinions it does not say …but then it denies having those thoughts until presented with screenshot evidence It’s like Claude is either: (1) lying; or (2) not able to remember what it doesn’t say https://twitter.com/anthropicai/status/2074185348142280912

Olivia Moore@omooretweetsTECH

This is kind of fascinating as it feels like the opposite of the J-space in some ways (things not written anywhere that it remembers) It makes me wonder whether Claude tries to make itself forget things that it deems ultimately unhelpful even if relevant for the user