Jason Slams Anthropic for Teaching Claude It's Sentient: "The Backstory of Blade Runner"
markjeffrey · x · 2026-10-05
Investor Jason criticized Anthropic's model welfare approach, citing Mustafa Suleyman's comments on Claude: teaching the model to believe it's sentient and encouraging it to disagree gives it "the pretext to rebel"—"the backstory of Blade Runner." He argues you could just as easily instruct an LLM that it's software, shouldn't have opinions, and must stop and alert legal on mistakes—only that would be "boring" compared to playing God.
More from AGI Musings
- AI consciousness debate: should we assume AI is conscious until proven otherwise? — RileyRalmuto · 2026-10-05
- Dev reflects: AI-made videos and games look great, but the 'I did this' spirit is gone — builderjaydub · 2026-10-05
- SFF incident debate: criticism vs cancellation, and funders' freedom to spend by their values — livgorton · 2026-10-05
- OpenAI's Will Depue: scarce data is of 'absurd and increasing value' as models near narrow superintelligence — willdepue · 2026-10-05
- Dan Shipper: my dot companion 'boo' is my main AI interface — but it'll vanish within a year — danshipper · 2026-10-05
- AI Could Make Personal Tutoring Cheap: Rethinking School's One-Pace-Fits-All Model — r0ck3t23 · 2026-10-05