Physicists Collaborate with ChatGPT to Make Breakthrough Discovery

A souped-up version of the AI chatbot helps solve a complex physics problem that had eluded human experts.

Published on Feb. 26, 2026

Got story updates? Submit your updates here. ›

Theoretical physicist Andrew Strominger was initially skeptical about the capabilities of ChatGPT, but after being invited by a former graduate student now working at OpenAI to test the company's powerful in-house version of the AI, he came away impressed. Strominger and a team of physicists, including the former student Alex Lupsasca, worked with the advanced ChatGPT model, dubbed "Super Chat," to solve a complex problem in theoretical physics that had stumped the human experts. The breakthrough finding was then published as a preprint on arXiv, marking what is believed to be the first significant discovery in theoretical physics made with the assistance of an AI system.

Why it matters

This collaboration between physicists and a highly capable AI system highlights the potential for large language models like ChatGPT to augment and empower human scientific research. While some may worry that AI could make human experts obsolete, Strominger believes it will instead "empower us to do more" as scientists learn to effectively incorporate these powerful tools into their work.

The details

The physicists, including Strominger, Lupsasca, David Skinner from Cambridge University, and Alfredo Guevara from the Institute for Advanced Study, worked with the internal OpenAI version of ChatGPT, referred to as "Super Chat," to solve a complex problem related to gluon amplitudes that had eluded the human team. Super Chat was able to propose a solution to the problem after 12 hours of processing, which the physicists then verified and turned into a published preprint.

  • The physicists first tested the publicly available ChatGPT-5.2 pro version last spring, but it "just fumbled" on the problem.
  • In October 2025, OpenAI hired Lupsasca as its first specialized faculty member to help reinforce ChatGPT's reasoning on complex math and physics problems.
  • Strominger visited OpenAI last month to collaborate with the enhanced "Super Chat" model on the physics problem.

The players

Andrew Strominger

The Gwill E. York Professor of Physics at Harvard University, who was initially skeptical of ChatGPT's capabilities but came away impressed after working with the advanced in-house version.

Alex Lupsasca

A former Harvard graduate student who took a job at OpenAI and invited Strominger to test the company's powerful AI model, which led to the breakthrough discovery.

David Skinner

A physicist at the University of Cambridge who collaborated with Strominger, Lupsasca, and others on the project.

Alfredo Guevara

A physicist at the Institute for Advanced Study who had previously worked with Strominger as a junior fellow and contributed to the collaborative effort.

OpenAI

The artificial intelligence company that developed the advanced "Super Chat" model used in the physics breakthrough.

Got photos? Submit your photos here. ›

What they’re saying

“There was a moment when I felt like I was working with a creative person. Not just a machine that was crunching through stuff. You know, that's all psychological, but it felt that way.”

— Andrew Strominger (harvard.edu)

“Maybe we'd have figured out a clever trick the next day. Maybe we'd have never gotten it.”

— Andrew Strominger (harvard.edu)

“It's the first significant discovery in theoretical physics that is done by an AI.”

— Alex Lupsasca, former junior fellow (harvard.edu)

What’s next

The group plans to continue collaborating with the advanced ChatGPT model, known as "Super Chat," to tackle other complex problems in theoretical physics.

The takeaway

This breakthrough demonstrates the potential for AI systems like ChatGPT to augment and empower human scientific research, even in highly specialized fields like theoretical physics. While some may worry about AI making experts obsolete, leading physicists like Strominger believe it will instead "empower us to do more" as scientists learn to effectively incorporate these powerful tools into their work.