Shibo Hao(郝世博)

Ph.D. student at UCSD

photo_5x4_cropped.png

Hello! I’m Shibo Hao, a Ph.D. student at UC San Diego, advised by Zhiting Hu. My research is funded by the Bloomberg Fellowship. I was a research scientist intern at Meta FIAR lab, mentored by Yuandong Tian and Jason Weston. I received my B.S. in Computer Science from Peking University.

My research goal is to push the boundaries of machine reasoning. My work includes training LLMs to reason with reinforcement learning (Guru, OREO, FoR), exploring reasoning in latent space (Coconut, Coconut-theory, Coconut-dynamics), building a system-2 reasoning framework using world-model planning (Reasoning via Planning, Pandora, LLM Reasoners), and augmenting LLMs with external tools (ToolkenGPT).

News

Sep 25, 2025 Guru, our exploration of cross-domain RL for LLM reasoning, and Reasoning-by-Superposition, A theoretical perspective on Coconut are accepted by NeurIPS 2025.
Apr 14, 2025 Coconut 🥥 is featured in Quanta Magazine!
Dec 21, 2024 Introducing OREO (Offline REasoning Optimization) (Arxiv, Twitter)
Dec 9, 2024 Honored to receive the Bloomberg Data Science Ph.D. Fellowship!
Jul 10, 2024 LLM Reasoners is accepted to the first Conference of Language Modeling (COLM 2024).
May 24, 2024 Check out Pandora, our new work towards a general world model 🌎
Nov 17, 2023 ToolkenGPT is accepted to NeurIPS 2023 as an oral presentation, and received the best paper award at SoCalNLP 2023🎉!
Oct 25, 2023 Reasoning via Planning (RAP) has been featured in State of AI Report 2023.

Talks

  • Reasoning beyond Imitating Human Language: Reinforcement Learning and Latent-space Reasoning [Slides]
    • Bloomberg Ph.D. Fellowship Virtual Symposium (May 2025)
    • Nvidia, Hosted by Boris Ginsburg (May 2025)
  • LLM Reasoners: New Evaluation, Library and Analysis of step-by-step reasoning with LLMs [Slides]

Selected publications

2025

  1. NeurIPS
    Reasoning by Superposition: A Theoretical Perspective on Chain of Continuous Thought
    Hanlin Zhu*, Shibo Hao*, Zhiting Hu, Jiantao Jiao, Stuart Russell, and Yuandong Tian
    Advances in Neural Information Processing Systems, 2025
    PDF Code
  2. ACL
    Offline Reinforcement Learning for LLM Multi-Step Reasoning
    Huaijie Wang*, Shibo Hao*, Hanze Dong, Shenao Zhang, Yilin Bao, Ziran Yang, and Yi Wu
    In Findings of the Association for Computational Linguistics: ACL 2025, 2025
    PDF Code
  3. NeurIPS
    Revisiting Reinforcement Learning for LLM Reasoning from A Cross-Domain Perspective
    Zhoujun Cheng*, Shibo Hao*, Tianyang Liu*, Fan Zhou, Yutao Xie, Feng Yao, Yuexin Bian, Yonghao Zhuang, Nilabjo Dey, Yuheng Zha, and  others
    Advances in Neural Information Processing Systems, 2025
    PDF Code
  4. COLM
    Training Large Language Models to Reason in a Continuous Latent Space
    Shibo Hao, Sainbayar Sukhbaatar, DiJia Su, Xian Li, Zhiting Hu, Jason Weston, and Yuandong Tian
    In Proceedings of the 2025 Conference on Language Models (COLM), 2025
    Featured by Quanta Magazine
    PDF Code
  5. Preprint
    Emergence of Superposition: Unveiling the Training Dynamics of Chain of Continuous Thought
    Hanlin Zhu, Shibo Hao, Zhiting Hu, Jiantao Jiao, Stuart Russell, and Yuandong Tian
    arXiv preprint arXiv:2509.23365, 2025
    PDF Code

2024

  1. Preprint
    Pandora: Towards General World Model with Natural Language Actions and Video States
    Jiannan Xiang*, Guangyi Liu*, Yi Gu*, Qiyue Gao, Yuting Ning, Yuheng Zha, Zeyu Feng, Tianhua Tao, Shibo Hao, Yemin Shi, Zhengzhong Liu, Eric P. Xing, and Zhiting Hu
    arXiv preprint arXiv:2406.09455, 2024
    HTML PDF Code
  2. COLM
    LLM Reasoners: New Evaluation, Library, and Analysis of Step-by-Step Reasoning with Large Language Models
    Shibo Hao*, Yi Gu*, Haotian Luo*, Tianyang Liu, Xiyan Shao, Xinyuan Wang, Shuhua Xie, Haodi Ma, Adithya Samavedhi, Qiyue Gao, Zhen Wang, and Zhiting Hu
    In Conference on Language Model (COLM), 2024
    Also to appear at Large Language Model (LLM) Agents workshop at ICLR 2024
    HTML PDF Code

2023

  1. NeurIPS
    ToolkenGPT: Augmenting Frozen Language Models with Massive Tools via Tool Embeddings
    Shibo Hao, Tianyang Liu, Zhen Wang, and Zhiting Hu
    Advances in Neural Information Processing Systems (oral), 2023
    Best Paper Award at SoCalNLP 2023
    PDF Code Poster Slides
  2. EMNLP
    Reasoning with language model is planning with world model
    Shibo Hao*, Yi Gu*, Haodi Ma, Joshua Jiahua Hong, Zhen Wang, Daisy Zhe Wang, and Zhiting Hu
    In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, 2023
    Featured in State of AI Report 2023
    Also to appear at GenPlan workshop at NeurIPS 2023
    PDF Code Slides