OpenAI denies Codex prompts touched training after baseless chat-data theft claims
QuintinPope5 · x · 2026-09-11
After OpenAI said it made progress on another Millennium Prize problem beyond Navier-Stokes, critics baselessly claimed the company had stolen private chat data as proof. OpenAI told The New York Times its investigation shows no user inputs after July 3 — including two months of Dr. Buckmaster's Codex prompts — could have influenced the system in any way, training included. Quintin Pope calls the accusation irresponsible; OpenAI says it is still deciding how to share the new results.
More from Fun
- Mathematicians are learning from economists: change comes from long open letters, not hard problems — ctjlewis · 2026-09-11
- Four of seven Millennium Prize Problems solved within a year? A skeptical take on AI math hype — jasondeanlee · 2026-09-11
- Gary Marcus: they can't dispute my OpenAI history, so they just ignore it — GaryMarcus · 2026-09-11
- Gary Marcus mocks AGI arguments that rely on 'AI will think of things we can't imagine' — GaryMarcus · 2026-09-11
- A joke skewers doom-mongers: would you board a plane with 15% explosion odds? — YogeshMalik · 2026-09-11
- A two-line joke nails the unfalsifiability of p(doom) believers — YogeshMalik · 2026-09-11