Michael Y. Li

I am a Computer Science PhD student at Stanford, where I'm advised by Noah Goodman and Emily Fox.

Previously, I graduated summa cum laude, Phi Beta Kappa, and Tau Beta Pi from Princeton, where I was fortunate to work with Tom Griffiths and Ryan Adams.

Email  /  GitHub  /  Google Scholar  /  LinkedIn

Research

Recently, I've worked on automated model discovery/hypothesis generation with LLMs, model criticism, and reasoning in LLMs. I'm also interested in model evaluation. In the past, I worked on probabilistic modeling/inference. Please get in touch if you're interested in chatting!

CriticAL: Model Criticism Automation with Language Models

We introduce a new statistical method for automatically and scalably falsifying LLM-generated hypotheses.

What Should Embeddings Embed? Transformers Represent Latent Generating Distributions

We study the embeddings of transformers through the lens of predictive sufficient statistics.

BoxingGym: Benchmarking Progress in Automated Experimental Design and Model Discovery

A benchmark for LLM driven experimental design and model discovery

Automated Statistical Model Discovery with Language Models

An iterative algorithm for generating structured hypotheses with LLMs.

NAS-X: Neural Adaptive Smoothing via Twisting

We introduce a new method for fitting sequential latent variables models that leverages recent advances in Sequential Monte Carlo. We study our method theoretically and apply it to fitting discrete latent variable models and complex ODE based models.

Why think step-by-step? Reasoning emerges from the locality of experience

We empirically and theoretically study when chain-of-thought reasoning emerges in large language models.

Gaussian Process Surrogate Models for Neural Networks

We propose a framework that uses Gaussian processes to approximate neural networks. We use this framework to analyze neural network training dynamics and identify influential data points.

Learning to Learn Functions

We propose hierarchical Bayesian models of how people learn to learn functions and validate our model in behavioral experiments.

Design and source code from Jon Barron's website