Alex L. Zhang

Hi! I broadly work on areas where language models are underutilized or inefficient. I am currently a 2nd year PhD student at MIT CSAIL advised by Omar Khattab and Tim Kraska. I am grateful to be a Laude Open Research Resident and a research fellow at Prime Intellect to push forward open science. I am also a recipient of the NSF GRFP.

To understand what I’m thinking about right now, read in order:

• Recursive Language Models (RLMs).
• The Mismanaged Geniuses Hypothesis.
• Empirical evidence of harnesses inducing locally in-distribution trajectories.
• Prime Agent, a production RLM harness.
• Speculative programmatic tool calling (PTC).

Some other notable things worth mentioning:

• I used to make a lot of benchmarks: VideoGameBench, KernelBench, SWE-bench Multimodal .
• I am part of the core team running the GPU MODE competitions and community.
• I graduated as the top student of the Princeton CS department.

I’ve had brief but pleasant stints at a variety of companies: Sakana AI, VantAI, Snap Research, Apple, and Claryo.

Feel free to reach out to talk about anything through my email at [x]@mit.edu where [x]=altzhang. I also help moderate the GPU MODE community. As a last fun fact, I used to make and sell PC games, one of which was mildly successful (~100k+ players). [unfinished example]