Gurbir Arora

I am a theoretical physicist with a Ph.D. in Physics from Brandeis University and am currently a Moore Postdoctoral Associate at Cornell University in the Earls lab. My research sits at the intersection of geometry, physics, and machine learning, with a particular focus on using geometric ideas to mechanistically interpret LLMs.

A big part of my Ph.D. was about understanding black holes, especially their interior. We can’t peek inside a black hole, but Hawking radiation offers a saving grace: it encodes information about the interior. My work studies interior reconstruction—how to recover interior data from exterior observations using a mix of analytical techniques and computer simulations. I built analytical tools and Python checks to verify when interior-reconstruction protocols work and how long reconstruction would take. Along the way, my collaborators and I found that the standard reconstruction protocol fails in certain spacetimes. We pointed this out in 2024, and we are finishing up work on a variant of the original Random Tensor Network that addresses the issue.

During undergrad at IIT Bombay, I explored a mix of problems: searching for dark-matter signatures in dwarf spheroidal galaxies with Fermi-LAT data; forecasting thunderstorms via time-series analysis of KUMODeS II measurements; studying the entanglement structure of AKLT models; and numerically analyzing the ringdown of a black hole. That last project earned me an Undergraduate Research Award, and I wrote an accessible introduction for Resonance.

Off-hours, I’m grinding Destiny 2 speedruns—tweaking routes, theorycrafting, and celebrating when a PB drops. I love learning about aviation—one of my favorite activities is plane spotting. I also cook a lot of Indian food, and I bike on weekends to reset.