afaik math problems at openai are being solved like this. i heard it on noam brown dwarkesh podcast