
OpenAI embarrassed themselves twice by passing off old solutions as GPT-5 discoveries
OpenAI managed to embarrass themselves twice on the same rake. And the second attempt came out even more embarrassing than the first.
Remember the hype about GPT-5 and the supposedly solved Erdős problem? Back then the model solemnly presented the solution to an open mathematical problem. Only that this solution was found in 20-year-old literature. Someone had solved the problem long ago, only everyone successfully forgot about it and continued considering it unsolved. The AI simply dug up a dusty publication, that’s all.
But these were still flowers. Now OpenAI announced that their model handled another 10 tasks from the same list, again found their solutions in forgotten works. And seems to be doing well, searches material quality, honestly cites sources. Normal search function, no breakthrough.
And now the most interesting part. OpenAI employees post about this with the wording: “GPT-5 found solutions to another 10 open math problems”. Somewhere in the comments they of course mentioned that solutions are taken from literature. But for anyone not aware of the backstory, this unambiguously reads as “AI discovered new mathematics”.
Twitter doesn’t forgive this. Instead of hype they got a wave of mockery and accusations. Even Nobel laureate Demis Hassabis from Google called it a disgrace. Information about discoveries in mathematics was deleted. And the authors from OpenAI justified themselves that they didn’t mean anything like that. There’s your marketing.