I am a working mathematician. A problem that I cared about greatly (and probably spent > 3000 hours working on) was on their list. I looked at the paper, and I have to say it is more clearly written than about 30% of the papers I typically referee. I don't want to name poorly written papers, but I agree that "there are plenty of published papers that are just as poorly written as OpenAI's".

And which papers do you typically referee? Without that information this claim is meaningless. For example, if you referee free-for-all papers that today are likely written by LLMs as well then sure I can understand that. But if you referee papers from grad students then that's more concerning.