Nice NYT article on AI and mathematics.

As impressive as natural language models are (ChatGPT, etc), it seems that a conceptual leap in design will be necessary before we see anything that looks like robust reasoning.