Hao (Jack) BAI

Hao (Jack) BAI

CS M.S. Candidate

University of Illinois, Urbana Champaign

Hi, there! I’m Jack. I’m an MSCS student at UIUC advised by Prof. Nan Jiang, and a visiting scholar at UC Berkeley advised by Prof. Sergey Levine. I’m fully devoted to autonomous intelligence upon two levels: (1) I study representation learning for foundation models in order to interpret their behaviors. (2) I study reinforcement learning algorithms for foudation models based on the so-derived understandings.

Previously, I was a visiting scholar of Prof. Yi Ma, and was a research assistant of Prof. Heng Ji and Prof. Chengxiang Zhai. Before master’s, I obtained a dual degree from Zhejiang University and UIUC in Computer Engineering. During those wonderful years, I interned at Microsoft Research (Asia) advised by Dr. Shilin He.

  • Representation Learning
  • Reinforcement Learning
  • MS in Computer Science, 2025

    University of Illinois, Urbana Champaign

  • BS in Computer Engineering, 2023

    University of Illinois, Urbana-Champaign


We have to learn the bitter lesson that building in how we think we think does not work in the long run. The two methods that seem to scale arbitrarily … are learning and search. Learning is to use data to extract patterns, and search is the optimization procedure that uses computation to make rational decisions.

Richard Sutton, The Bitter Lesson, 2019

Data without optimization doesn’t allow to solve new problems in new ways; optimization without data is hard to apply to the real world outside of simulators.

Sergey Levine, RL with Large Datasets, 2023


External candidates should self-identify or be referred to have strongly matched research taste with me and solid engineering skills.

  • [Fall 2023] Referred external candidates from UIUC or UC Berkeley will be considered.
  • [Spring 2024] Referred external candidates from UC Berkeley will be considered.
  • [Summer 2024] No external collaboration are considered during this period.

Core Publications

Fine-Tuning Large Vision-Language Models as Decision-Making Agents via Reinforcement Learning (Preprint)
White-Box Transformers via Sparse Rate Reduction: Compression Is All There Is? (JMLR)
CharmBana: Progressive Responses with Real-Time Internet Search for Knowledge-Powered Conversations (WSDM-24)
Social Commonsense-Guided Search Query Generation for Open-Domain Knowledge-Powered Conversations (EMNLP-23)