What are latent state models for training dynamics useful for? (1/N)
A Gaussian mixture model predicts convergence time about as well as the HMM it’s meant to be a strictly weaker version of.
Sep 20, 2026
William Dai
I’m a senior at Yale studying mathematics and computer science. I work on interpretability and cognitive science: what language models learn, and how we’d know.
In Yale’s Computational Linguistics lab that’s meant linguistic generalization — what models pick up about syntax, and how to trace it back to the data they were trained on.