OpenAI announces three big science discoveries and a new way to work with academics
The Verge Richard Lawler ● Covered by 2 sources
OpenAI says it found three more big science results in math, computer science, and physics. This time outside experts helped shape how it shared the work, not just the work itself.
Based on reporting by The Verge, Richard Lawler — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Less than a month after dumping hundreds of manuscripts online and saying they cracked long-running math problems, OpenAI is back with three more claims of “major” scientific progress. The new batch touches math, computer science, and physics. It’s the kind of announcement that wants to sound like a breakthrough parade, and OpenAI is clearly leaning into that tone again.
What’s different is the process around the results. For the computer science and physics work, OpenAI says external experts acted as “shepherds,” guiding the results. On the math side, it took a different tack: the company says those outside advisers told it to release the AI output so researchers could build their own explanations from it.
That’s a notable shift. OpenAI isn’t just publishing conclusions and moving on; it’s experimenting with how to hand scientific claims to other people and let them do something useful with them. The company also added the results to GitHub today, which fits the theme: less sealed-off lab mystique, more public artifact for others to inspect.
The first listed result is called Local smoothing for the Euclidean wave equation. The source says the proof uses an auxiliary par, but doesn’t go further there. Even so, the bigger story is the same one OpenAI keeps pushing: not just that its models can spit out ideas, but that they can be wrapped in a workflow that makes academics part of the process.
My take — AI-written commentary, not fact-checked reporting
This is the sensible version of AI science theater: put humans in the loop before the hype machine declares victory. OpenAI’s move to use “shepherds” is a quiet admission that raw model output isn’t science by itself, no matter how shiny the announcement. That’s progress, but it also shows how much of this field still runs on presentation first and proof second.
Read more about this at: The Verge
