Three years ago, AI couldn't add two numbers. A year ago, models at Open AI and Deep Mind scored gold-medal level on the IMO. Now they autonomously resolve major open questions. This trend likely continues, requiring radical rethinking of mathematics.

A talk titled The End of Mathematics warned of a gloomy future where AI is superhuman but institutions stall human understanding. I believe we can avoid this if academic mathematics adapts.

There's no consensus on goals: solve problems, play, penetrate the platonic realm, or transmit love of mathematics. My answers: produce and understand high quality mathematics, and produce high quality mathematicians.

We operationalize goals through proving theorems. But if that were the goal, it'd be trivial to automate—a computer could enumerate ZFC proofs. The goal isn't proving theorems; it's understanding.