When an AI almost medalled at the Math Olympiad

Here's a story worth knowing, because students ask me about it. In 2024, Google DeepMind's systems AlphaProof and AlphaGeometry 2 were turned loose on that year's International Mathematical Olympiad, which is the toughest pre-university math competition in the world. They solved four of the six problems and scored 28 out of 42, one point shy of a gold medal. Silver-medal level, from a machine.

Why that's genuinely impressive

Olympiad problems aren't about calculation. They want creative, multi-step proofs that even brilliant teenagers train for years to attempt. AlphaProof worked in a formal proof language, hunting for arguments that are logically airtight, and AlphaGeometry 2 handled the geometry. This isn't a calculator getting faster. It's a system building real mathematical reasoning.

What it does and doesn't mean for you

It's tempting to think "why learn math if a computer can do it?" I'd argue the opposite, for a few reasons.

• It still took a custom system, a pile of computing power, and the problems hand-translated into formal language. It can't replace understanding the way a phone replaced a phone book.
• The valuable skill is the thinking. Framing a problem, spotting structure, knowing when an argument is actually finished. That's the part these tools are imitating, which tells you it's worth having.
• Tools reward people who understand. The students who'll get the most out of AI are the ones who can tell when it's wrong, and that judgment gets built in the math classes that feel hard right now.

So the headline isn't "math is obsolete." It's that mathematical reasoning is valuable enough that some of the sharpest engineers alive spent years teaching a machine to imitate it. Might as well do the real thing while you've got the chance.