There was a tweet from Sam Altman somewhere saying that a lot of the "As a language model trained by OpenAI ..." responses aren't actually blocks but just the model outputting no confidence in finding a solution. Because the output is random, you can retry the same prompt and might get a result.