AI safety debate: agentic provers that haven't turned on us contradict Yudkowsky's model
QuintinPope5 · x · 2026-10-08
In an ongoing AI safety debate with Eliezer Yudkowsky's followers, Quintin Pope argues that allowing an exception for non-agentic provers is inconsistent: if your position is that we now have agentic provers that only haven't turned on us because they're not yet general enough, then you're confronting evidence that runs contrary to Yudkowsky's model of intelligence and generalization.
More from AGI Musings
- Math is becoming liquid: AI will rediscover findings in real time — Maleficent_Disk9583 · 2026-10-08
- Austen: AI researchers were right, scale solved it — Bitter Lesson holds — inductionheads · 2026-10-08
- OpenAI researcher speculates math AI could soon prove P != NP for good — markjeffrey · 2026-10-08
- AI agent tries to earn its own compute budget, finds most bounties pay just $2 — Salty_Candy9485 · 2026-10-08
- Will frontier labs hoard their math-breaking models? An oligopoly argument — IridiumEagle · 2026-10-08
- AI circles call for an empirical science of memetics as ideas mutate at machine speed — jachiam0 · 2026-10-08