What's up, I’m Max, and I want AI to go well for everyone! I keep this broad on purpose, so I can follow the problems I think matter most. Right now I’m a MATS fellow in Berkeley, mentored by Prof. Stephen Casper.
I am an undergrad at UW–Madison studying Computer Science and Data Science and graduating in the fall of 2026. There, I am part of a student group full of ambitious, capable people I've learned a lot from (WAISI). My research has spanned transferable adversarial materials with UChicago XLab, entity-level hallucination detection with Prof. Sharon Li, and most recently a project I led on how company choices shape video-deepfake abuse.
I was born and raised in southern California. I have grown up swimming, playing waterpolo, and enjoying the ocean. I like to read, eat/cook good food, work out, and travel. Most of all, I love meeting new people! So please reach out if you'd like to chat.
A small number of open-weight video models have become the dominant tools for non-consensual deepfake generation. We trace how developer and distributor choices — data curation, safeguards, moderation — foreseeably shape downstream misuse, and argue that risk management at the model and platform level can substantially reduce harm.
Uncertainty-based hallucination detection usually works at the sentence or paragraph level. We introduce HalluEntity — a benchmark of 18,785 entity-level annotations — and evaluate detection methods across 17 modern LLMs, finding token-level approaches over-predict while context-aware methods do better but remain suboptimal.
Retained as an expert consultant for some lawsuits on AI product safety and the foreseeable misuse of generative image/video models, drawing on my deepfake-abuse research.
Selected for a competitive AI-policy accelerator preparing people for careers shaping how AI is governed.
Built a multi-agent system from scratch showing how jailbreaks cascade between agents, and presented it to congressional staffers in D.C. with the Wisconsin AI Safety Initiative (WAISI).
This summer I’m a MATS 10.0 fellow in Berkeley, working on technical AI governance and improving unlearning + tamper resistance for open-weight models in dual use domains.
Alongside that, I’m consulting on some lawsuits involving AI product safety and the foreseeable misuse of generative image/video models, exploring the bay area, and trying to make as many friends as possible.