Jiangjie Chen (陈江捷) is a researcher at ByteDance Seed Team. In 2024, he earned his Ph.D. at Fudan University in the School of Computer Science, Shanghai, China. His current interested research topics are mostly around building reasoning models and autonomous agents:
Ph.D. in CS, 2019 - 2024
Fudan University
B.S. in CS (honors), 2014 - 2019
Fudan University
May. 2025: Check out Enigmata! We propose a comprehensive suite of puzzles for improving logical reasoning of reasoning models, tailored for RLVR training. We find that not only do such puzzles drastically improve puzzle reasoning of LLMs, but also improve SoTA models such as Seed1.5-Thinking on challenging reasoning tasks such as AIME and GPQA! This is a free lunch for SoTA models, since Enigmata is synthetic and can be generated at scale!
May. 2025: I was awarded with Nomination Award for Outstanding Doctoral Dissertation of Shanghai Computer Society!
May. 2025: Our papers DEEPER and HistoryAnalogy are accepted to ACL 2025!
May. 2025: Our paper CoSER is accepted to ICML 2025! Check out this comprehensive resource for role-playing agents!
Apr. 2025: Presenting Seed-Thinking-v1.5 from ByteDance Seed Team, a cutting-edge reasoning model that’s incredible in math, code, science, and logical reasoning!
Mar. 2025: DAPO is out! A new critic-free RL algorithm that directly trains a pre-trained base model to SoTA performance on AIME 2024 without any SFT.
Mar. 2025: Four papers accepted to NAACL 2025: SelfGoal, EvoAgent, EasyTool and Barrier in Language Agent Planning.
Oct. 2024: Three papers accepted to NeurIPS 2024 Workshop on Open-World Agents: EvoAgent, SelfGoal and AucArena. See you in Vancouver!
Sep. 2024: Our survey paper on role-playing agents is accepted to TMLR!
Sep. 2024: We have three accepted papers in EMNLP 2024! Two main papers are Segment+ on long-context processing with short-context models, and CROSS on role-playing evaluation, and one finding paper DetectBench on benchmarking detective reasoning.
Jul. 2024: Our work on Irrelevant Evidence got accepted in COLM 2024!
Jul. 2024: I have graduated from Fudan University, and will officially join ByteDance Seed Team as a Full-time researcher.
Jun. 2024: How to automatically extend the specialized agent to multi-agent systems to improve task-solving capability? We propose EvoAgent, a generic method to automatically extend expert agents to multi-agent systems via the evolutionary algorithm. EvoAgent can be generalized to any LLM-based agent framework, and significantly enhance the task-solving capabilities of LLM-based agents!
Jun. 2024: Want your agents to win an auction for you? But does your agent know what it means by such a vague and high-level goal as “winning an auction”? Check out SelfGoal! We propose an automatic approach that enhances language agents’ capabilities to achieve high-level goals with limited instructions and delayed feedback by adaptively breaking down goals into practical subgoals. Really excited about automating agents to do high-level task with minimal human instructions!
Jun. 2024: TravelPlanner got a Spotlight recommendation at ICML 2024!
May 2024: Just defended my thesis, officially a Dr. :)
May 2024: Four papers are accepted to the main conference of ACL 2024! They are: TimeArena, AnalogyKB, InCharacter and GumbelSoft! See you in Bangkok :)
May 2024: Our TravelPlanner got accepted to ICML 2024!
We introduce Enigmata, the first comprehensive suite tailored for improving LLMs with puzzle reasoning skills.
We introduce Seed-Thinking-v1.5, a Mixture-of-Experts (MoE) model with a relatively small size, featuring 20B activated and 200B total parameters, capable of reasoning through thinking before responding, resulting in improved performance on a widerange of benchmarks.
We introduce DAPO, a Decoupled Clip and Dynamic sAmpling Policy Optimization algorithm, and fully open-source a state-of-the-art large-scale RL system that achieves 50 points on AIME 2024 using Qwen2.5-32B base model.
We introduce TravelAgent, an LLM-powered travel planning system that generates rational, comprehensive, and personalized itineraries through four modules, demonstrating effectiveness in dynamic scenarios.
We introduced TravelPlanner, a benchmark for assessing language agents’ planning abilities, showing that even advanced models like GPT-4 face difficulties with complex tasks.
We present the first comprehensive and controlled investigation into the behavior of large language models when encountering knowledge conflicts.