Three frontier lab CEOs agree on slowing AI — a researcher lays out why implementation is brutally hard
Yuchenj_UW · x · 2026-09-13
Yuchen Jin (YuchenjUW) notes it's striking that Dario Amodei, Elon Musk, and Sam Altman all now agree AI progress should slow. While broadly supportive of embedded third-party evaluators like METR, he raises three concerns: who evaluates the evaluators and keeps them fair and unbiased; slowing progress conflicts with frontier labs' incentives, especially ahead of IPOs; and you can only pace what you can measure — how do you define and measure something like recursive self-improvement? His verdict: the idea sounds good, but implementation seems extremely hard.
More from AGI Musings
- Beyond Slowdown vs. Acceleration: A 'Group C' Case for AI Safety Without Personal Attacks — tristanbob · 2026-09-13
- Dario's Essay a 'Good First Step' but Still Vague, Says Pacing Details of AI Agreement Need Scrutiny — jam3scampbell · 2026-09-13
- 'Verifiers Need Verification': Skepticism Grows Over Independence of AI's Proposed Independent Evaluators — _arohan_ · 2026-09-13
- AI Could Actually Lower GDP: ChatGPT Queries Barely Counted While Plumber Work Is, Argues Economist — moultano · 2026-09-13
- AI may actually shrink GDP: count real output, not dollars — moultano · 2026-09-13
- Naval: AI won't slow down — the only real choice is centralize or disperse power — TinfoilTricorn · 2026-09-13