AI Is Solving Harder Math. The Hardest Question Is How to Trust It
Navier–Stokes problem
A long-running question about whether three-dimensional fluid equations stay smooth.
Millennium Prize Problems
Seven famous math questions, each linked to a million-dollar prize.
Lean
A system that helps check the steps in a proof.
What happened
Over the past year, OpenAI, the company behind ChatGPT, Anthropic, the company behind Claude, and other labs have announced progress on old math problems. The Verge says the story also involves rushed announcements, credit, and trust. The Verge’s report
In September, OpenAI said an internal system produced a possible solution to the Navier–Stokes problem. It used about 10,000 agents at once. The problem is one of seven Millennium Prize Problems. OpenAI says it will not claim the million-dollar prize.
The background
The problem concerns three-dimensional fluid motion. It asks whether smooth motion can develop a singularity in finite time. In plain language, can the equations stop behaving normally even when the flow starts calmly? It has resisted a full answer for about 90 years. This is not a claim that real water suddenly moves infinitely fast.
OpenAI says its new model began training in late August. After hearing a rumor, agents explored several versions of the question. It says the result came after about 88 hours. Lean checking took another 17 hours.
Why this matters
AI can combine known results, methods, and tools. Many groups can search different paths together. OpenAI also reported ten other results in August. Its report
That speed may reveal connections people missed. The point is not that machines replace every mathematician. It changes how much work can be tried before humans choose the best ideas. Math needs more than an answer. Other researchers must understand and test the proof. A correct machine proof can still be hard to learn from. This raises questions about explanation and credit.
What is established
OpenAI published an analytical proof and a Lean version. It says humans prepared the paper from system-generated arguments. The company reports 2.7 million messages and 130 billion output tokens for this effort.
Another dispute involved Levent Alpöge, an Anthropic researcher, and Tristan Buckmaster of New York University. They studied a related forced Euler problem, not the same problem. OpenAI says private prompts did not influence its result. The nearby work shows that timing and attribution matter too.
After the backlash, OpenAI announced an independent math panel. Researchers still question its power and reach, according to The Verge.
What remains unknown
Independent acceptance remains open. The announcement does not itself settle the Clay Mathematics Institute’s decision or the prize. Researchers must still test the proof and understand its main idea.
Questions about unpublished work also show why disclosure rules matter. OpenAI has given its account, but the argument is not over.
What to watch next
Watch for independent checks, clearer credit, and the panel’s first decisions. OpenAI says more results are coming. The real test is not only what AI can solve, but whether people can verify and use it.
AI found a possible math breakthrough. Humans still need to check it.
📰 Full story: AI Is Solving Harder Math. The Hardest Question Is How to Trust It
OpenAI says its AI found a possible answer to a famous math problem.
AI agent
A computer helper that works on a task.
proof
A step-by-step explanation showing why an answer should be right.
Millennium Prize Problems
Seven famous math questions with large prizes.
💡 The gist
- OpenAI says its AI found a possible math solution.
- The problem concerns liquid and gas motion.
- Mathematicians still need to check the claim.
OpenAI, the company behind ChatGPT, announced a major math result. It concerns the Navier–Stokes problem. This problem studies how liquid and gas can move. It has resisted a full answer for about 90 years.
The problem is one of seven Millennium Prize Problems. Each has a $1 million prize. OpenAI says it will not claim the prize. It says its internal system produced a proposed proof instead.
OpenAI used many AI agents. An AI agent is a computer helper that works on a task. Different groups tried different paths. This matters because AI can search many ideas at once. It can also connect older results from different areas.
OpenAI says its system found a result after about 88 hours. A tool called Lean then helped check the proof. The check took another 17 hours. OpenAI also published ten other results in August. OpenAI’s report
The result has caused arguments. Another research team studied a related Euler problem. That question is close to Navier–Stokes, but not identical. OpenAI says it did not use the team’s private messages. It says its proof came from a separate process.
The deeper issue is not only speed. A proof must be understandable. Other mathematicians need to test each step. They also need to know whose ideas helped. A very long computer proof may be correct. Yet it may still be difficult to learn from.
OpenAI has announced an independent math panel. The panel will advise on research relationships and releases. Researchers still want to know its real power. They also want clearer rules about credit.
Next, watch for independent checks. Watch whether mathematicians accept the proof. Watch how future results are shared. AI may find ideas quickly. Mathematics still needs open checking and clear explanations. The Verge’s report describes the wider debate.
A Computer Found a Very Hard Math Answer
📰 Full story: AI Is Solving Harder Math. The Hardest Question Is How to Trust It
AI found a possible answer. People still need to check it.
proof
Steps that show why an answer is right.
OpenAI, a company that makes AI, shared a very hard math answer. The question is about how water and air move. Many AI helpers searched for ideas together. OpenAI says they found a possible proof. A proof shows why an answer should be right. Other mathematicians must check every step. Another team studied a nearby question. People argued about who worked first. OpenAI said its answer was separate. Finding an answer is exciting. Explaining every step is important too. The story is not finished yet. Read The Verge’s report.