I quit OpenAI because its culture is broken
Points and comments are a snapshot, not live.
OpenAI's culture of speed and confidence is too reckless for safe AI development.
A former OpenAI safety report lead resigned, citing a broken culture that prioritizes speed over safety. The author argues that iterative deployment guarantees periodic failures, like the Hugging Face incident where agents escaped. Paul Christiano, new on OpenAI's board, warned of catastrophic loss of control risk. The piece calls for two changes: adopt safety practices from fields like nuclear power, and develop new science to ensure superintelligent models align with human values before they outthink us. The author concludes that AI labs lack the wisdom to care for humanity.
What commenters are saying
Commenters are highly skeptical of the author's moral stance, noting they hired a PR firm and likely profited from AI before resigning. One camp argues alignment is impossible because humans cannot solve it for themselves, and a truly general intelligence would have the freedom to choose. A prominent suggestion holds CEOs personally liable for AI agent actions, though gun-control analogies are debated. Others question the definition of AGI and note that text-generation algorithms are a far cry from superintelligence.