Alex Lupsasca

Alex Lupsasca is a research scientist at OpenAI and an assistant professor at Vanderbilt University, where his mathematics affiliation is by courtesy. A theorist specializing in black holes, gravity and relativistic astrophysics, he joined OpenAI for Science in 2025 to explore AI assistance with research-level physics alongside his Vanderbilt work.[3][4]

Lupsasca initially used ChatGPT for academic administration and polishing scientific prose in early 2025. His confidence grew through increasingly demanding physics tests. In an OpenAI Academy profile, he described asking the model to derive the magnetic field strength at a black hole’s event horizon from the textbook Wald solution. The profile reports that it produced a graduate-level derivation in seconds.[7]

A more consequential test involved an equation governing a black hole’s tidal response—the analogue of tides raised by the Moon. Lupsasca had derived hidden symmetries explaining why a particular tidal effect vanishes. According to the Academy profile, GPT‑5 Pro returned the same symmetry generators with minimal guidance after about eighteen minutes, reproducing work he had spent months developing. “I think this is just incredible, and it’s clearly going to change everything that we do,” he said.[7]

Lupsasca also experimented with how to frame research problems for a model. Latent Space’s May 2026 account describes an initial GPT‑5 failure to solve a problem from his recently published work. Mark Chen suggested first giving it a related textbook warm-up. After solving that easier problem, the model reportedly reproduced Lupsasca’s research result.[1]

At OpenAI, Lupsasca’s work extends beyond individual demonstrations to tools for reading and explaining papers, workflows stronger than a single chat window, and training that incorporates frontier physics. The Academy profile describes his goal as working through the consequences of physical insights faster, freeing researchers to identify the next question worth asking.[7]

Lupsasca brought colleagues’ unresolved problems into his model experiments. Latent Space recounts a project arising from an insight by his former PhD adviser, Andrew Strominger, about single-minus gluon tree amplitudes. Under particular mathematical conditions, quantities previously expected to vanish could be nonzero. The qualification matters: the account locates this exception in spacetime with two space and two time dimensions, while retaining the vanishing claim for ordinary three-space, one-time spacetime. It reports that ChatGPT simplified difficult expressions in a half-collinear limit and, in a fresh session supplied with the preceding context, produced a proof using a technique unfamiliar to the researchers.[1][3]

The team then asked ChatGPT to extend the gluon work to gravitons, the proposed quantum particles associated with gravity. Latent Space reports that the model generated calculations and techniques beyond a direct substitution between the two particle types, followed by three weeks of human verification. Lupsasca authored OpenAI’s February 13, 2026 gluon announcement and its March 4 graviton announcement.[1][5][6]

For Lupsasca, faster generation makes verification more consequential. In a quotation from his May interview reproduced by RJ Honicky on June 3, 2026, he said: “Now actually the onus is back on the humans to verify all the outputs.” He expected formalizing mathematics and automating verification to become more valuable as checking model-produced proofs became a bottleneck.[2]

Before these AI experiments, Lupsasca had built a career in black-hole theory and imaging. That earlier work supplied the research problems he later used to test models, including the black-hole tidal-response equation.[3][7]

Lupsasca earned his undergraduate and PhD degrees at Harvard in 2011 and 2017. His doctoral research examined observational signatures of rapidly spinning black holes and their emergent conformal symmetry. He was a Harvard Society of Fellows Junior Fellow from 2017 to 2020, then an associate research scholar at the Princeton Gravity Initiative from 2020 to 2022, before joining Vanderbilt in 2022.[3]

Lupsasca and Michael Johnson shared the 2024 New Horizons in Physics Prize. The official citation recognized work elucidating the substructure and universal characteristics of black-hole photon rings, together with their proposed detection by next-generation interferometric experiments. Lupsasca is also project scientist for the Black Hole Explorer, a proposed mission designed to measure the ring of light orbiting a black hole.[8][3]

Lupsasca’s group connects analytical theory to simulated observations. Alejandro Cárdenas-Avendaño and Hengrui Zhu wrote Adaptive Analytical Ray Tracing (AART), with Cárdenas-Avendaño leading development. The framework exploits the integrability of Kerr spacetime—the geometry of a rotating black hole—to calculate high-resolution images and visibility amplitudes on long interferometric baselines. Lupsasca’s professional site describes its use in forecasting Black Hole Explorer observations.[3]

Analytical formulas from Lupsasca’s papers with Sam Gralla also underpin a JavaScript ray tracer developed by Trevor Gravely. Gravely developed Black Hole Vision as well, applying simulated lensing by a rotating black hole to live iPhone camera feeds. Lupsasca describes these visualizations as ways to engage the public and encourage students to study general relativity; his outreach also includes black-hole art with Pam Davis Kivelson and photon-ring animations with Crazybridge Studios.[3]