Yu Hu (胡宇)
PhD Student
IntelliArchLab
Rice University
Houston, TX

yh188@rice.edu
github.com/yu44MTX

Hi, I’m Yu.

I am a first-year PhD student at Rice University, working in IntelliArchLab.

My research focuses on diffusion models, deep-supervised LLMs, and AI security — with a current emphasis on multi-turn jailbreak attacks against aligned language models.

What I am thinking about

Three threads I am pulling on in parallel:

• Diffusion model efficiency. Dynamic-resolution sampling, trajectory shaping, and what makes “near-the-end” denoising steps so quality-critical.
• Deep-supervised LLMs (DS-LLM). Reasoning models with an internal orchestrator, rather than scaffolding-by-prompt from the outside.
• AI security and red-teaming. Multi-turn attack frameworks that probe alignment, using diffusion-style LLMs as the attacker.

A longer version of these threads, with project context, lives on the CV page.

Writing

I write here on two cadences:

• Notes. Short, self-contained derivations and reading notes — usually math or code.
• Essays. Occasional longer pieces about ideas that wouldn’t fit into a paper.

Recent posts:

• 2025-10-30 — The Normal Distribution
• 2025-04-16 — Monkeys vs Apes
• 2024-10-04 — Iterators vs Generators in Python

The full archive is on the Blog page.

Elsewhere

The best way to reach me is by email. Code lives on GitHub.