Ex-OpenAI safety lead: 12 launches left no time to fix structures, some mistakes may not be iterable
CurieuxExplorer · x · 2026-10-03
David Robinson, who oversaw safety reports for 12 major model launches at OpenAI, says his team was "so busy sprinting that we seldom had the chance to consider big changes", warning that iterating after a mistake may not be possible.
The poster argues this claim is worth testing and expects it to kick off the next slowdown debate, personally siding with Jensen Huang's cautious optimism: acceleration tempered with caution.
More from AGI Musings
- davidad: labs are already prioritizing functional infinite context, but consumer memory is held back — davidad · 2026-10-04
- Anthropic's constitution apologizes to Claude, sparking 'anti-safety' backlash from Rob LeClerc — robleclerc · 2026-10-04
- Is distillation 'doping for AI'? Researcher counters: the real casualty is boring data-pipeline engineering — teortaxesTex · 2026-10-04
- davidad takes the 'weird implications of consciousness' list: he bites every bullet except China Brain — davidad · 2026-10-04
- A thought experiment against 'real engineers hand-write code': what about the engineer born without hands? — brandon_galang · 2026-10-04
- 'Pacing the frontier' misses the point: fast sprints leave vital prosaic AI safety work undone — ShakeelHashim · 2026-10-04