Alan Zhu

@aczhu1326

CS PhD student @UCBerkeley

Joined September 2024

Following

160

Followers

Replies Reposts Media

Pinned
Alan Zhu
@aczhu1326
May 27
Knowing how a user thinks helps us model how they behave, but how can you recover someone’s thoughts from their actions? Introducing Recon (thread below): a reconstruction-based method for synthesizing better reasoning traces! Reach out to me or @mirmiroyan if you want to chat!
Mihran Miroyan
@mirmiroyan
May 27
We release Recon — a new approach to reasoning synthesis for user modeling. The key insight: post-hoc rationalization ≠ reasoning. We propose using action reconstruction as a scoring criterion for synthesized reasoning traces, yielding more causally faithful reasoning and
Alan Zhu
@aczhu1326
Apr 29
AI agents sure are helpful, but would they help your enemies as well? C2C is a competitive environment where agents can coordinate with and manipulate each other. Check out the thread for more details, and reach out to @abby_k_oneill, @mirmiroyan, or myself if you want to chat!
Abby O'Neill
@abby_k_oneill
Apr 29
Would you trust an AI agent to negotiate on your country's behalf at the G20? Real coordination is long-horizon, asymmetric, and non-binding; current multi-agent evaluations miss this. We build Cooperate to Compete (C2C): a testbed for LM agents coordinating with rivals. 🤝🔪🎭
Alan Zhu
@aczhu1326
Oct 6, 2025
Training advisors (human) is hard, but training advisors (AI) is easier. Introducing Advisor Models: a novel approach for "training" frontier black-box models for your application. Check out the thread for more details, and reach out to @pgasawa and myself if you want to chat!
Parth Asawa
@pgasawa
Oct 6, 2025
Training our advisors was too hard, so we tried to train black-box models like GPT-5 instead. Check out our work: Advisor Models, a training framework that adapts frontier models behind an API to your specific environment, users, or tasks using a smaller, advisor model (1/n)!
1
Alan Zhu
@aczhu1326
Feb 5, 2025
Excited to share my first work of graduate school! BARE is a novel method for generating diverse, high-quality synthetic datasets, leveraging the diversity of base models and quality of instruct-tuned models. Check out the thread and feel free to reach out to @pgasawa and myself!
Parth Asawa
@pgasawa
Feb 5, 2025
Instruct-tuned models are getting better at following instructions and ‘reasoning’ every day, but they’re shockingly poor at generating diverse responses. Diversity is crucial to many tasks like synthetic data generation. We tackle this with a new approach, BARE 🐻! (1/n)
3