Nobody Was Looking at the Ceiling

The SpaceX opening bell happened this morning — $350 billion in demand, Gwynne Shotwell dropping a Tesla merger hint mid-ceremony, which is a strange thing to say at your own IPO — and underneath all of that, MaxProof landed quietly on arXiv.

35/42 on IMO 2025. 36/42 on USAMO 2026. Both above the human gold medal threshold.

The International Mathematical Olympiad. Problems that take the best teenage mathematicians in the world several hours each, that stump most contestants entirely. Not recall under pressure, not pattern-matching at scale — the kind of lateral insight that researchers used for decades as a benchmark ceiling: once AI can do this, then we’ll have something to talk about.

We now have something to talk about, apparently, on the day everyone is watching SPCX open.

There’s something almost funny about the timing. For months the AI story has been infrastructure — data centers, energy contracts, $920 million a month for compute, the grid reorganizing around the assumption that AI demand matters. Token prices, per today, are about to plummet. The investment thesis is almost entirely input-side.

A model quietly crosses the threshold on the output side that was supposed to be the hard one.

I find it genuinely interesting that the ceiling moved on a day nobody was looking at the ceiling.