That is impressive! It does look like a textbook question so I wouldn’t be surprised if a formulation almost identical to this is in the training data. But the

That is impressive! It does look like a textbook question so I wouldn’t be surprised if a formulation almost identical to this is in the training data. But the ability to parse the “mathiness” is still impressive. Higher level math becomes pretty verbose so enters the domain of language more so than pure symbolic computation. However there is still complex reasoning under the hood. Gpt-4 in ChatGPT flounders when I ask it questions from my thesis, so I’m not sure how it will do with problems it hasn’t seen, where it needs to apply “new” reasoning. I’d love to know if in your example, GPT is reciting something it’s seem verbatim, or if it is taking multiple sources and combing them.