Max Zhuang

Hi, I’m Max!

I’m an undergraduate at UC Berkeley studying Molecular & Cell Biology and Data Science, and I’m most interested in how models can be made useful for biology and medicine, from understanding how cells work to designing new drugs.

This site is how I keep track of what I’m reading, building, and trying to understand. Lately, I've been thinking a lot more (and working on) connecting protein language models with reinforcement learning - especially how we can do it in a way that utilizes the data efficiently. This site houses a few things: projects I've been working on, short digests that talk about interesting fields of research, and rankings/ratings of some papers I've read recently.

More about meSay hello

Working on

All projects

Antibody Steering

I’m testing what post-training does to an antibody model that edits a binding loop, asking which objectives improve ranking, which collapse the model’s samples while the headline number still rises, and what it takes to get the gain without the collapse.

In progressCode on GitHub
Binding data and parent replay train the model that writes the loop.

Digests

All digests
  • 53 min read

    Evaluating mid-training

    What it means for one mid-training recipe to be better than another, why a mid-trained checkpoint's own scores mislead in specific ways, and how to compare recipes so that the comparison holds.

  • 33 min read

    Combining unreliable evidence

    What rule a trained network applies when it combines several inputs that can each be wrong, why it does not double count dependence that stayed fixed throughout its training, what it takes to learn to discount a copy, and what language models, majority votes and label aggregators do with evidence that is not independent.

  • 24 min read

    Training a latent variable instead of searching for it

    When a trained network can stand in for a per-input search over an unobserved quantity, such as a move sequence, a chain of thought, an autoencoder's code or a 3D scene, what it loses, where search still has to come back, and how game players and reasoning models combine the two.

  • 29 min read

    Transfer from tasks with exact answers

    When training a language model on problems with checkable answers, such as math, code and puzzles, makes it better or worse at other tasks; what the exact answer contributes; and why the task being measured decides more than the task being trained.

Recently read

All readings