Alignment Circle Debate: Are Frontier Labs Overconfident in Goal-Shaping?
jeremygillen1 · x · 2026-08-19
Alignment researchers Jeremy Gillen and Jacques Thibs continue sparring. Thibs argues labs haven't really internalized how bad they are at shaping goals and expect it will just be solved "enough"; on uncooperativeness, some don't believe it, others think monitoring will keep AIs sufficiently cooperative. Gillen counters that if they expect this to be solved, they shouldn't advocate for better sandboxing.
More from AGI Musings
- Robots get cheaper every generation; humans expect raises — VraserX · 2026-08-19
- Zvi comments on 'user-centric AI': rejecting corporate ideological imposition — TheZvi · 2026-08-19
- Rich Sutton: Synthetic Data Is a Mistake, LLMs Are Only a Quarter of Intelligence — GregCook2011 · 2026-08-19
- NVIDIA health lead: AI automates tasks, not jobs—and clinician demand is rising — nvidia · 2026-08-19
- "The Moon We Made": a concept trailer imagining a tightly regulated AI future — mvult · 2026-08-19
- AI hyperscalers' $308B debt buildout is pushing up Treasury yields — tszzl · 2026-08-19