1h ago

Oxford's Toby Ord argues AI math breakthrough costs are miscalculated by ignoring the cumulative expense of failed trials

Solving a FrontierMath hypergraph problem cost under $100,000.

——0——
Original post
OPToby Ord#1159Toby Ord@TOBYORDOXFORD

It would be great to know how much it currently costs per major mathematical research result from AI. People keep using the API cost for a successful try, but forget to add the API costs for all the unsuccessful ones! By that logic, it only takes me one arrow to shoot a bullseye.

3:12 AM · May 27, 2026 View on X
REPLYToby Ord#1159Toby Ord@TOBYORDOXFORD

Obviously these things will get cheaper, but it is very useful to understand the current costs.

Toby OrdToby Ord@tobyordoxford

At the moment, I think the current cost per big result that the mathematics community cares about is greater than a lifetime's salary for a top mathematician.

10:20 AM · May 27, 2026 · 733 Views
10:20 AM · May 27, 2026 · 367 Views
REPLYJaime Sevilla#1389Jaime Sevilla@JSEVILLAMOL

@tobyordoxford I don't think the world spent collectively even $100k on finding a solution to the Ramsey Hypergraphs FrontierMath problem.

epoch.ai
A Ramsey-style Problem on Hypergraphs
Construct hypergraphs as large as possible that do not have a certain easy-to-check, difficult-to-find property.
Toby OrdToby Ord@tobyordoxford

At the moment, I think the current cost per big result that the mathematics community cares about is greater than a lifetime's salary for a top mathematician.

10:20 AM · May 27, 2026 · 733 Views
11:06 AM · May 27, 2026 · 28 Views