about

Zizhao Hu

Los Angeles

CS PhD @ USC · GLAMOR Lab

advised by Jesse Thomason & Mohammad Rostami @ Amazon

CV scholar

“Context is the new weight.”

I'm a researcher and engineer on Agentic AI. I work on agentic memory, synthetic data training, and self-improving AI.

I earned my BS in Physics at Georgia Tech and am now a CS PhD at USC.

I'm conducting LLM unlearning research for the US Government's IARPA. Previously I was an ML domain lead at Handshake AI and a data engineer at Scale AI, where I collaborated with teams from OpenAI, Meta, and Anthropic to improve their unreleased black-box models.

what i work on

memory

Agentic Memory

Continual learning of AI agents — in-context learning, continual fine-tuning, and unlearning.

world model

World Model

In-context world models, adaptation to post-training task worlds, and adapting agents in evolving envs.

latency

Low-Latency AI

Efficient attention architectures, KV-cache compression, latent segmentation, and recurrent transformers.

safety

AI Safety

Synthetic data training, risks of multi-agent interaction, post-training guardrails, and AI behavioral study.

news & media

2026-05· preprint
arXiv preprint: SHRED — Document Unlearning via Self-Distillation and Entropy Demotion
2026-03· preprint
arXiv preprint: Expert Personas Improve LLM Alignment but Damage Accuracy — Bootstrapping Intent-Based Persona Routing with PRISM
2026-03· coverage
Media coverage on PRISM paper — The Register, AIToday, Tencent News, 36Kr, QbitAI, Yahoo Tech
2025-12· fellowship
Wrapped up Project Canary
2025-10· talks
Presented Multimodal Synthetic Data Finetuning and Model Collapse at ACM ICMI
2025-08· fellowship
Joined Project Canary (Handshake AI)
2025-07· fellowship
Started Handshake AI Fellowship 2025

my path

2016
Physics
photonics & metasurface design · dynamic systems
2018
Agile Systems
bio-inspired flight, sensing, and locomotion
2021
Robotics · RL
policy learning for physical control and agent behavior
2022
Continual Learning · VAE
regularization design for variational autoencoders
2023
Multimodal Generation
diffusion models · vision-language model architecture
2025
Multiagent
coordination, division of labor, and mutual verification across agents — continual adaptation at the population level
2026
Agentic Memory
in-context learning, continual fine-tuning, unlearning, and memory scaffolds — adapting agents at the context and model level
next
World Models
predictive world models and the architectures to serve them in real time