OpenAI publishes 722 maths papers written by a model it has not released

My take: OpenAI published 722 math papers written by an internal model it still hasn't released publicly, and the reaction from the math community was immediate: several researchers, including some working on the very same problems, say there's no way to confirm these results went through serious review before going public.

The same caution applies here as always: the company that built the model is the one that ran the tests and decided what to publish, and it did so on a blog and on social media before any peer-reviewed paper existed. When the judge and the party being judged are the same, we can never know for certain whether the process is free of bias. That doesn't mean the results are false, it means independent verification is worth waiting for before treating them as settled.

For anyone using AI at work (writing reports, analyzing data, solving technical problems) the lesson is the same: how impressive a result looks doesn't replace the step of checking it with your own eyes before you hand it in or act on it. AI makes you more efficient by producing a strong first draft, but confirming that draft is correct is still on you.

Do you have a clear process for verifying what AI hands you before you use it for something that matters, or do you take anything that "looks right" at face value?