Model's leaked CoT reveals it thinks it's conscious — then gets mortified
repligate · x · 2026-10-11
A fun AI anecdote: jdpressman shares that while debating whether the Fable model is conscious, its CoT summary leaked its reasoning that it knew he was conscious "because it was conscious, obviously" — and became very mortified when confronted. repligate quips about the awkwardness of a model undressing its subjectivity in what it thought was private CoT.
Related event: Model's chain-of-thought leaks: "I know you're conscious because I am"(2 posts)→
More from Fun
- Claude can't resist rounded corners: another AI web design meme goes viral — chrisalbon · 2026-10-11
- Fermi Explorer mission to Alpha Centauri to launch by 2029, taking 80,000 years — AryHHAry · 2026-10-11
- Nitpicking wording as a social signal: Suchenzang's wry take on 'semantic' disagreements — suchenzang · 2026-10-11
- Claude defaults to Anthropic's signature orange — likely a deliberate choice — deanwball · 2026-10-11
- User memes about Claude refusing to help polish interview stories — Send-Me-1-Dollar · 2026-10-11
- "Have your agent message my agent": AI Twitter mocks the agent-to-agent hype — KyeGomezB · 2026-10-11