AI researcher argues new model results undercut the strongest recent 'doom' argument
jd_pressman · x · 2026-11-21
- jdpressman argues that the strongest doom argument of recent years — Jeremy Gillen and peterbarnett's Jan 2024 LessWrong report — is significantly undermined by recent model results.
- The report's logic: AIs capable of large-scale novel science will be approximately consequentialist; behavior-based training leaves goal specification underdetermined; alignment is itself out-of-distribution, so humanity must solve it or face catastrophe. The author sees current results as weakening both this argument and the broader case for doom.
Related event: Gillen-Barnett doom argument weakened by newest models, researcher says(2 posts)→
More from AGI Musings
- beh_zod's optimistic take on tech job market: jobs are rebundled, not lost — beh_zod · 2026-11-21
- Gary Marcus: No next funding wave in sight, many AI companies face bankruptcy — GaryMarcus · 2026-11-21
- Percy Liang: you can believe AI is transformational and that bottlenecks persist — RishiBommasani · 2026-11-21
- Google research scientist: LLMs have made me smarter than at any point in my career — docmilanfar · 2026-11-21
- AI safety debate: capability, power and loss of control are three different things — random_walker · 2026-11-21
- One-liner: AI could mechanomorphize humanity until replacing us counts as saving us — PierceLilholt · 2026-11-21