Nathan Henry

cat about_me.txt

• I’m a PhD student in Computer Science at UC Berkeley, advised by Stuart Russell. I’m broadly interested in AI safety and in making AI go well. I previously completed the Engineering Science program at the University of Toronto.
• This semester I’m a TA for Scalable AI (EE 290/194), a graduate course at UC Berkeley covering the full LLM lifecycle, including performance engineering, pre-training, post-training, efficient inference, safety, deployment, and agents.

ls ~/current-research-interests

• pre-training safety
• interpretability
• using AI to understand AI
• understanding how and why neural networks generalize
• AI for mathematics

ls ~/outside-of-research

• I have three dogs, with two cats on the way.
• I have an Erdős–Bacon number of 6.
• I like learning about languages (a little Mandarin and ASL, fluent in neither).
• I love spicy food and trying dishes I haven’t eaten before.

ls ~/selected-papers

• Transformers Learn Two Competing Sorting Algorithms, and Only One Length-Generalizes. Khashayar Gatmiry*, Nathan W. Henry*, Chloe H. Su, Emmanuel Abbe, Samy Bengio, Melanie Weber, Jonathan A. Kelner, Peter L. Bartlett. Preprint forthcoming. *equal contribution.
• Geometry of Lightning Self-Attention: Identifiability and Dimension. Nathan W. Henry, Giovanni Luca Marchetti, Kathlén Kohn. ICLR 2025. [arXiv]