It's a very important clarification if it took $2000/problem on 20 problem attempts or on 1,000 problem attempts for each successful one. That may be the deciding factor on whether or not it's economically viable to replace a mathematician with a ChatGPT subscription.

No, this is valid criticism. Oai gives the impression anybody could get similar results at a similar price, but that’s very likely not true. This is marketing first, then mathematics.