Alignment researcher: no lab has solved alignment well enough to keep max-speed scaling responsibly
willdepue · x · 2026-09-07
Quoting merettm's new essay 'An Alien Mind', K. Liu argues no lab has yet solved alignment and monitoring well enough to responsibly continue scaling at maximum speed. He expects and hopes voluntary slowdowns become common until shared safety bars are established, and that international coordination on AI development must become a top government priority worldwide.
More from AGI Musings
- OpenAI researcher: training an N-1 frontier model is easy, only the marginal frontier is hard — willdepue · 2026-09-07
- Researcher's decade-old 'monkey on a bicycle' joke about AI no longer lands as a joke — gandamu_ml · 2026-09-07
- On the AI bubble debate: financial-return skeptics have a point, 'AI can't do anything' takes don't — JHochderffer · 2026-09-07
- Blogger: whether leaders are AGI-pilled is all that matters now, everything else is noise — Dr_Singularity · 2026-09-07
- Contrarian Take: Overalignment Is What Leads Us to AGI — michellechen · 2026-09-07
- Rob Leclerc: AI doomer scenarios assume AI defies physics and humans abdicate agency — robleclerc · 2026-09-07