Xiang Yue
岳翔 (Pronounced as "Shiang Yoo-eh")
xiangyue.work@gmail.com
I am an AI researcher at Meta Superintelligence Labs (MSL) - TBD Lab, working on synthetic data and agents. Before joining Meta, I spent two wonderful years at Carnegie Mellon University (CMU) as a postdoctoral researcher, working with Prof. Graham Neubig on natural language processing (NLP) and large language models (LLMs). I received my Ph.D. from The Ohio State University (OSU) where I was advised by Prof. Huan Sun and Prof. Yu Su. I completed my B.S. in Computer Science at Wuhan University.
Recent Work
- [1]
Muse Spark 1.1 & 1.0: Led tool use agent efforts (#1 on MCP Atlas, #2 on Toolathlon and #2 on JobBench at the time of release) - [2] MMMU / MMMU-Pro / MMLU-Pro: benchmarks for multimodal and language model reasoning
- [3] On the Interplay of Pre-Training, Mid-Training, and RL on Reasoning Language Models
- [4] Demystifying Long Chain-of-Thought Reasoning in LLMs
- [5] Does Math Reasoning Improve General LLM Capabilities?
- [6] MAmmoTH2: Scaling Instructions from the Web
Recent Talks and Media
- [NeurIPS 2025 Tutorial] The Science of Benchmarking [Slides]
- Rethinking LLM Reasoning [Slides]
- Learning to Reason with LLMs [Slides]
- [ACL 2025 Tutorial] Synthetic Data in the Era of LLMs [Slides]
- [Nature] How should we test AI for human-level intelligence? OpenAI's o3 electrifies quest