OpenAI's credibility takes two hits in one day

OpenAI withdrew three mathematical results, and the thread's first reaction was that the write-ups "still read like slop" and felt rushed. The sharpest question was whether the withdrawn proofs were Lean-checked at all. A companion post, "OpenAI, the Partition Principle, and Mathematics", came from a mathematician who found the AI-written proofs hard to enjoy or even follow. Defenders replied that the models will improve and that experts' complaints just become the next training set.

In a separate thread, OpenAI fired three safety researchers for "mishandling research information." One says she was fired for accessing an executive's email, access she says OpenAI itself delegated to her for recruiting. The BBC headline says they were let go for "prioritising safety." Commenters tied this to a belief that OpenAI is under competitive pressure and that safety people are the first friction to go.

The pattern here: a lab can be right about capability and still lose on process. Unverified output and ambiguous firings both make people ask who is checking the work. A commenter's worry about "unsupervised maths" shipping a bug in the foundations is the same worry as a safety team being let go.