What do AI's scientific discoveries actually give us?

A claimed proof that a famous assumption is wrong, new results in physics, and the honest answer to "ok, but what do I get out of it?"

This week OpenAI said its models produced new results in three corners of science. A claimed disproof of the Exponential Time Hypothesis, a big assumption computer scientists use to say how fast certain problems can ever be solved. A proof about how waves behave, with ties to long-standing geometry problems. And existence results in quantum field theory, the maths behind how particles interact.

Impressive on paper. Independent experts still have to confirm all of it, so treat it as claimed, not settled.

Fine. But what do I actually get out of this? Cheaper things, cures, faster software? Or is this just something to stare at in awe?

Straight answer: right now, nothing you can touch. Disproving that hypothesis would not make your apps faster, and it would not settle the famous P versus NP question. The physics result is not the million dollar prize problem either, because that one also needs a piece the reported result does not cover. Awe, yes. Groceries, not yet.

What does change, quietly

The bottleneck moves. When a machine can produce a proof, the scarce thing is no longer having the idea, it is checking it. The maths here comes with formal proofs that software can verify. But software only checks that the argument holds for the statement it was given, not that the statement captures the problem humans cared about. That judgement still needs experts, time and money. A new foundation fund of 15 million dollars, with contributions from both OpenAI and Anthropic, exists exactly to help mathematicians cope with this.

And there is a fight about access. The maths proofs are public. The physics group is holding back the AI-generated materials until it writes its own explanation, and independent mathematicians say that is not how open science works. Whoever sees the evidence first gets to build on it first.

The one that might touch your life sooner

Medicine looks closer to the ground. A diagnostic system published in a top journal reached 83 percent on hard published cases, against 36 percent for doctors with books and internet and 22 percent without. Caveat in capital letters: those are retrospective tests, not real patients. A hospital study is next.

So the honest story is not "AI cured something". It is that the way we confirm truth is about to be stress-tested harder than it ever was.