OpenAI GPT-5.2 Pro Helps Physicists Crack Quantum Gravity Problem

OpenAI's GPT-5.2 Pro has helped a team of physicists from Harvard, Cambridge, and the Institute for Advanced Study derive new mathematical results in quantum gravity, according to a preprint published March 4, 2026. The AI model not only solved the core problem but produced a preliminary draft of the research paper itself.

The work extends findings from a February 2026 paper on gluon amplitudes to gravitons—the theoretical quantum particles of gravity. Both papers overturn decades of textbook assumptions that certain particle interactions, called single-minus amplitudes, must equal zero at tree level.

What the researchers actually found

Scattering amplitudes calculate the probability of particle interactions. Physicists have long assumed that when one graviton has negative helicity while all others have positive helicity, the resulting amplitude vanishes under standard approximations.

Wrong, apparently. The new preprint demonstrates these amplitudes exist as well-defined mathematical distributions when particle momenta align in what's called the half-collinear regime. The authors—Alfredo Guevara, Alexandru Lupsasca, David Skinner, Andrew Strominger, and Kevin Weil—derived explicit formulas describing these interactions.

The result connects to an infinite-dimensional "w-(1+∞)" symmetry that Roger Penrose identified in classical gravity half a century ago. Many physicists believe this symmetry holds the key to reconciling quantum mechanics with Einstein's general relativity. The preprint shows how this symmetry acts on gravitons for the first time.

How GPT-5.2 Pro contributed

After completing the earlier gluon paper, researchers fed it to GPT-5.2 Pro as context. They then asked the model to construct corresponding amplitudes for quantum gravity—work that would have taken human physicists considerably longer to derive manually.

The model solved the problem using the directed matrix-tree theorem, a technique the authors described as "beautiful and surprising." All results were subsequently verified analytically and checked against known physical limits using conventional methods.

OpenAI published a full transcript of the initial exchange between researchers and GPT-5.2 Pro, showing the model's step-by-step derivation process.

A shift in how physics gets done

The research team noted something telling about their workflow. Most time between the February gluon result and this graviton paper was spent confirming derivations, checking consistency, and preparing formal write-ups—not generating initial conjectures.

That's a meaningful inversion. Verification and exposition now consume the dominant share of effort, while AI handles the creative mathematical leaps. The transition from gluons to gravitons took weeks rather than the months or years such extensions typically require.

The authors are already investigating further extensions. For AI-focused investors and observers, this represents concrete evidence that frontier language models can participate meaningfully in theoretical research while maintaining rigorous scientific standards—a capability that could reshape R&D timelines across multiple industries.