Why self-invited AI evals won't work: PwC-style oversight and the China chip angle
xiaosun86 · x · 2026-09-13
A four-point take on frontier AI oversight: self-invited evals won't work for OpenAI and Anthropic (like using your phone's screen-time check on your own X usage); a PwC/KPMC-style gov-business structure sharing responsibility is more likely; China will follow suit with its own training chips (inference already usable, Apple seeking Chinese DRAM, "Dario is clueless on manufacturing"); and it all hinges on who first delivers real-world impact by curing a disease like ALS — AI's upside outweighs the fear.
More from AGI Musings
- Intelligence Has a Speed Limit: control theory caps recursive self-improvement — docmilanfar · 2026-09-13
- AI weapons/factory veteran: AI safety is for their benefit, not yours — tawnniee · 2026-09-13
- "AGI has essentially arrived, just not publicly": reports of existential crises at OpenAI and Anthropic — Neurogence · 2026-09-13
- Stolen compute could briefly undercut rogue-agent economics, then dry up — voooooogel · 2026-09-13
- Biggest AI risk comes from laziness and the human need for belonging, argues founder — arieljalali · 2026-09-13
- DeepMind's Alex Irpan doubles his p(doom) estimate from 2% to 4% — AlexIrpan · 2026-09-13