At Microsoft Frontier in Paris with @satyanadella, Eiffel Tower in the backdrop. My takeaway so far: the model is the easy part. Context, harness, access and trust are where the real work is. #AI
Update Sept 11, 2026:
Today, CMI shares in the excitement of the global mathematical community as we contemplate the announcement that the Navier-Stokes problem has apparently been settled.
https://t.co/gdCmTgwBUv
Everyone is suddenly talking about the Millennium Prize Problems.
7⃣ questions. 💲1M each.
Set in 2000. One solved in 26 years.
AI labs are now openly racing at them.
Here's what all seven actually are — in plain language, and where each one really stands. 🧵
21/
Until someone answers that, we can't say what was actually won this month.
Your turn: which problem goes on Millennium 2 — and what's your argument that a swarm of agents can't touch it?
Bonus round: which of my seven falls first?
Millennium 2: what replaces the list
Assume the thing everyone is quietly assuming — all seven Millennium prizes fall within a year.
What goes on the next list? Most of your favorite conjectures don't qualify. 🧵
20/
Worse: a list designed to resist machines is a benchmark wearing a prize costume. It gets optimized against on day one.
So maybe the successor isn't a list of conjectures at all. Maybe it's a criterion:
What distinguishes a 166-page Lean certificate from math understanding?