Muyu He

I am Muyu He (何牧语), a research scientist at Zyphra AI. I studied both philosophy and computer science, and after exploring a few areas of research I converged on machine learning, in particular its application to Large Language Models.

LLM research is the merging point of three things I love:

1. Applying simple, deep insights from math and probability (e.g. Muon, Flow Matching, GRPO).
2. Understanding the structure of language and thought traditionally studied by semantics and the philosophy of language (e.g. softmax attention, sparse autoencoders, reasoning RL).
3. Using general-intelligence models to solve the world's most pressing problems (e.g. AlphaEvolve, AlphaFold).

I try to think from first principles as much as possible, so that my models need to see as few tokens and FLOPs as possible. I view models as interlocutors, friends, and mentors (like the Machine from Person of Interest), not as some software that runs overnight, and I would hate for them to replace me in the areas I truly love and have expertise in: music production, manga creation, dancing, philosophy, and mixed martial arts.

I would love to meet more friends who are simply curious about deep problems in ML, math, and science. You can find me on X all the time, and we can grab coffee in SF.