About

I’m a senior at Yale studying mathematics and computer science. I came in off competition math — USAMO in 2022 — expecting to do pure math, and got stuck on a linguistics question: language models clearly learn something about syntax, but it’s surprisingly hard to say what, why, or how.

These days, I’m mostly interested in the intersection of cognitive science and AI interpretability. Cognitive scientists have spent decades studying the original black-box computational system, the human brain, and they still don’t quite understand it. Interpretability researchers have complete access to the model. They also still don’t quite understand it. Maybe having access was never the bottleneck; maybe it was what counts as explanation in the first place. Whatever the answer is, I’d expect it to hold up from every side: the data going in, the internals doing the work, and the behavior that they add up to.

In practice that’s meant working in Yale’s Computational Linguistics lab on linguistic generalization experiments – what’s in the learner’s input and how data shapes generalization behavior.

Away from research I like to read books that are about themselves — Pale Fire, House of Leaves.

Publications

What Exactly do Children Receive in Language Acquisition? A Case Study on CHILDES with Automated Detection of Filler-Gap Dependencies Zhenghao Herbert Zhou, William Dai, Maya Viswanathan, Simon Charlow, R. Thomas McCoy, and Robert Frank. Proceedings of the 30th Conference on Computational Natural Language Learning (CoNLL), 2026, pages 692–706. [arXiv] [Code]

Contact

The best way to reach me is william.dai@yale.edu. I answer email from people I don’t know.