Researchers at DeepMind and Anthropic: no viable scientific plan exists for recursively self-improving AI risks
sjgadler · x · 2026-09-10
A researcher who worked at Google DeepMind and now Anthropic, speaking in personal capacity, says a common sentiment among peers is that "there is not yet a viable scientific plan to solve risks from recursively self-improving AI."
It quotes a resigned Anthropic pretraining researcher who spent three years across OpenAI and Anthropic, accusing both companies of "racing straight to self-improving superintelligence and gambling with our lives."
Related event: Frontier Lab Alignment Leads Admit No Plan for Superintelligence(3 posts)→
More from AGI Musings
- Commenter: anyone regulating AI should prove they understand how it works — intellectronica · 2026-09-10
- Flipping doomer logic: pausing AI means 'advocating mass death', argues viral post — inductionheads · 2026-09-10
- AI safety researcher says the safety crowd was vindicated, but too credulous about corporate commitments — S_OhEigeartaigh · 2026-09-10
- Ben Todd on why AI hasn't boosted GDP yet: revenue 3x/year, acceleration a couple years out — connoraxiotes · 2026-09-10
- Calls to pause AI are a PSYOP to crush open-source labs, argues dev — JFPuget · 2026-09-10
- You can vibe code a feature, not a brand: Brinker on AI's organizational limits — chiefmartec · 2026-09-10