PeopleofSuperintelligence

Researcher

Damai Dai

Also known as 代达劢

Damai Dai is a language-model researcher whose work spans live-video generation, Transformer knowledge and sparse model design. He coauthored LiveBot at Peking University and later studied knowledge neurons and in-context learning; one of his papers won EMNLP’s 2023 Best Long Paper Award. He completed a PhD at Peking University in 2024 and coauthored DeepSeekMoE and DeepSeek-V3, followed by an analysis of DeepSeek-V3 and R1 systems. As of October 2026, he is a DeepSeek researcher and infrastructure lead, and his work includes a conditional-memory method for language models.

Now at
DeepSeek
Works
77
Citations
4,407
h-index
24
Badges

Publication numbers from OpenAlex, as of 19 October 2026.

Timeline

  1. 2019

    Generating live-video comments

    In 2019, he coauthors LiveBot at Peking University, a neural system that generates live comments from video and text context.

  2. 2022

    Mapping knowledge neurons

    In 2022, he coauthors research that introduces knowledge neurons to locate factual knowledge in pretrained Transformers and explores how those neurons can edit facts without fine-tuning.

  3. 2023

    Explaining in-context learning

    In 2023, he and his coauthors explain GPT’s in-context learning as implicit fine-tuning and connect Transformer attention to gradient descent.

  4. 2023

    An EMNLP paper award

    In 2023, “Label Words are Anchors,” a paper he coauthors, wins EMNLP’s Best Long Paper Award; it studies how label words guide in-context learning.

  5. 2024

    A PhD and sparse models

    In 2024, he earns a PhD at Peking University’s School of Computer Science under Professor Zhifang Sui at the MOE Key Lab of Computational Linguistics. He coauthors DeepSeekMoE, which uses fine-grained expert segmentation and shared experts, and the DeepSeek-V3 technical report on a 671B-parameter model with 37B activated per token.

  6. 2025

    Analyzing DeepSeek’s systems

    By 2025, he is part of DeepSeek’s research team as a recent Peking University doctoral graduate. In May 2025, he coauthors an analysis of the DeepSeek-V3/R1 architecture and AI infrastructure.

  7. As of October 2026

    DeepSeek infrastructure lead

    As of October 2026, he is a DeepSeek researcher and infrastructure lead responsible for inference-system engineering and large-scale deployment. In 2026, he coauthors work that introduces conditional memory through Deep Sparse Embedding, using local n-grams for retrieval.

Career

  1. Since 2021DeepSeekResearcher and infrastructure lead

Education

  1. 2024Peking UniversityPhD

From the biography and OpenAlex affiliations (the years of papers that list the organization), so some years are approximate.

Most-cited works

  1. DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning2025 · Nature · 1,112 citationsWith Bingxuan Wang, Bochao Wu, Chengqi Deng, Daya Guo and Zhibin Gou
  2. A Survey on In-context Learning2024 · Conference on Empirical Methods in Natural Language Processing (EMNLP) · 636 citations
  3. Can Open Large Language Models Catch Vulnerabilities?2025 · 538 citationsWith Bingxuan Wang, Bochao Wu, Chengqi Deng and Zhibin Gou
  4. DeepSeek-V3 Technical Report2024 · 292 citationsWith Bingxuan Wang, Bochao Wu, Chengqi Deng, Daya Guo and Dejian Yang
  5. DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language Models2024 · Annual Meeting of the Association for Computational Linguistics (ACL) · 282 citationsWith Chengqi Deng and Zhenda Xie

From OpenAlex, as of 19 October 2026.

  1. The median h-index is 53 at universities and 19 at frontier AI labsIncludes Damai Dai. The median h-index is 53 at universities and 19 at frontier AI labs. We compare publishing careers, citations and moves to estimate how universities and frontier labs contribute to AI research.
  2. The company share of AI researchers’ top papers rose from 2% to 32%Includes DeepSeek. We estimate where leading AI researchers worked when their most-cited papers appeared. The company share rose from 2% in 1975–1979 to 32% in 2020–2024.
  3. Google has the highest combined h-index among AI organizationsIncludes DeepSeek. Our estimates of combined and median h-index at AI organizations, using current affiliations and OpenAlex records.
  4. Google alumni founded or led 87 other organizations, the most among the labs we compareIncludes DeepSeek. We trace 212 organizations founded or led by AI lab alumni, and 223 career moves between the labs.

Colleagues

People with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Damai Dai yet. Send us a free-licensed photo.

  1. LiveBot: Generating Live Video Comments Based on Visual and Textual Contexts | Proceedings of the AAAI Conference on Artificial Intelligence ojs.aaai.org. Checked 19 October 2026.
  2. Knowledge Neurons in Pretrained Transformers - ACL Anthology aclanthology.org. Checked 19 October 2026.
  3. Why Can GPT Learn In-Context? Language Models Secretly Perform Gradient Descent as Meta-Optimizers - ACL Anthology aclanthology.org. Checked 19 October 2026.
  4. 计算机学院孙栩团队荣获EMNLP最佳论文奖-北京大学计算机学院 cs.pku.edu.cn. Checked 19 October 2026.
  5. About - Lean Wang’s Homepage leanwang326.github.io. Checked 19 October 2026.
  6. Damai Dai — China AI Atlas ai.techbuzzchina.com. Checked 19 October 2026.
  7. DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language Models - ACL Anthology aclanthology.org. Checked 19 October 2026.
  8. DeepSeek-V3 Technical Report arxiv.org. Checked 19 October 2026.
  9. China’s AI disrupter DeepSeek bets on low-key team of ‘young geniuses’ to beat US giants tech.yahoo.com. Checked 19 October 2026.
  10. Insights into DeepSeek-V3: Scaling Challenges and Reflections on Hardware for AI Architectures arxiv.org. Checked 19 October 2026.
  11. Conditional Memory via Scalable Lookup: A New Axis of Sparsity for Large Language Models - ACL Anthology aclanthology.org. Checked 19 October 2026.