Yang Zhilin (born November 1992) is a Chinese artificial intelligence researcher and entrepreneur who co-founded Moonshot AI in 2023 and became its chief executive officer. His research includes neural language modelling and language understanding, notably the jointly developed Transformer-XL and XLNet models. (zgcforum.com.cn)
Education and research career
Yang is from Shantou, Guangdong, China. He received his bachelor’s degree from Tsinghua University in 2015, where his adviser was Jie Tang. In 2019, he earned a PhD in Language and Information Technology at Carnegie Mellon University, supervised by Ruslan Salakhutdinov and William W. Cohen. The university’s alumni record identifies him as a graduate of its Language Technologies Institute. (zgcforum.com.cn)
During his time in the United States, Yang also conducted artificial intelligence research at Google Brain and Meta. His publications address topics including natural language processing, language-model pretraining, reading comprehension, and learning from limited labelled data. (zgcforum.com.cn)
Research contributions
Transformer-XL
Yang and Zihang Dai were equal-contribution lead authors of the 2019 paper introducing Transformer-XL. The model addresses a limitation of the conventional Transformer architecture: processing text in separate, fixed-length segments can restrict the dependencies a language model learns across segment boundaries. (kimiyoung.github.io)
Transformer-XL combines segment-level recurrence, which reuses hidden representations from preceding segments, with a relative positional encoding scheme. These mechanisms allow information to carry across segments while preserving temporal relationships. The paper evaluated the model on several language-modelling benchmarks and reported improvements in modelling long-range dependencies and in evaluation efficiency under its experimental conditions. (arxiv.org)
XLNet
Yang and Dai were also equal-contribution lead authors of XLNet, introduced in 2019. XLNet uses a generalized autoregressive pretraining objective that considers different orders in which token probabilities are factorized. This enables the model to learn from bidirectional context without using the same masked-token prediction procedure as BERT. It also incorporates mechanisms from Transformer-XL. (kimiyoung.github.io)
The original XLNet paper reported results on language-understanding tasks including question answering, sentiment analysis, and natural-language inference. Those results describe the models, datasets, and evaluation settings used in that study rather than a permanent ranking of language-model methods. (arxiv.org)
HotpotQA
Yang was an equal-contribution lead author of the 2018 paper introducing HotpotQA, a question-answering dataset designed to test reasoning across multiple documents. Its questions require combining evidence rather than locating an answer in a single passage. The dataset also supplies sentence-level supporting facts, allowing evaluation of both answers and the evidence used to obtain them. (kimiyoung.github.io)
Moonshot AI
In 2023, Yang co-founded Moonshot AI, a Chinese company developing large language models and the Kimi assistant. A 2026 Zhongguancun Forum biography identifies him as the company’s founder and chief executive officer. (zgcforum.com.cn)
Kimi Chat was launched in 2023. In an interview recorded that November, shortly after its launch, Yang discussed the company’s approach to model development and the organizational demands of artificial intelligence research. His entrepreneurial work followed his academic contributions to language-model architecture, pretraining, and question-answering datasets. (about.geekpark.net)
References
- 杨植麟zgcforum.com.cn
- Zhilin Yangkimiyoung.github.io
- Zhilin Yang — Language Technologies Institute — Carnegie Mellon Universitylti.cs.cmu.edu
- Transformer-XL: Attentive Language Models Beyond a Fixed-Length Contextarxiv.org
- XLNet: Generalized Autoregressive Pretraining for Language Understandingarxiv.org
- HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answeringarxiv.org
- HotpotQA Homepagehotpotqa.github.io
- The faces behind China's AI rise: Meet the founders of DeepSeek, Z.ai and Moonshot AIchannelnewsasia.com
- “Different Organizations Allow Different Things to Emerge”: Yang Zhilin’s First In-Depth Interviewabout.geekpark.net