From Futurism · Frank Landymore · · 1m 30s
OpenAI's math win turns into a theft fight
OpenAI said 10,000 agents cracked a Millennium Problem, then a mathematician said the work was his.
In brief
OpenAI said 10,000 agents cracked a Millennium Problem, then a mathematician said the work was his. When a lab can flood any promising research direction with agents, the safest move for a researcher becomes silence, and that breaks open science. Originally reported by Futurism.
A million dollar equation
The Navier-Stokes equations describe how fluids move, from wind to ocean currents. The Clay Mathematics Institute put them on its Millennium Problems list in 2000, with $1 million for a solution.
10,000 agents, 88 hours
OpenAI says it set 10,000 AI agents loose on the problem for 88 hours at a cost of millions in compute, according to head of research Mark Chen.
They found a blowup
The agents surfaced a scenario where the equations describe fluid accelerating to infinite velocity, a singularity. Physically impossible, which means the equations have a breaking point.
Someone got there first
Hours before the announcement, NYU mathematician Tristan Buckmaster said he had been on this for years, was close to publishing, and that OpenAI's path looked suspiciously like his own.
The Codex problem
Buckmaster had used OpenAI's Codex to check his work. OpenAI denied seeing it but admitted it "cannot rule out" that de-identified data from his usage improved its models.
“cannot rule out that de-identified data derived from their usage of our products helped improve our models”
Why would you ruin your career
Buckmaster also says OpenAI's Sebastien Bubeck pushed him to drop co-author Levent Alpoge, who works at Anthropic. Bubeck calls the account false and inflammatory.
“Why would you ruin your career?”
The headline
When a lab can flood any promising research direction with agents, the safest move for a researcher becomes silence, and that breaks open science.
