PeopleofSuperintelligence

Researcher

Lijun Yu

Also known as 于力军

Lijun Yu is a Staff Research Scientist at Google DeepMind who works on Gemini and on multimodal foundation models, especially video generation. After studying computer science and economics at Peking University, he went to Carnegie Mellon, where Alexander Hauptmann advised him as a master's student and where he earned a Ph.D. in 2024. He was first author of MAGVIT, a masked video transformer that became a CVPR 2023 Highlight, and of "Language Model Beats Diffusion," whose MAGVIT-v2 tokenizer lets language models outperform diffusion models on ImageNet and Kinetics. He co-authored VideoPoet, a language model for video generation introduced in December 2023, which received a Best Paper award at ICML 2024.

Badges

Timeline

  1. By 2021

    Peking University and GamesMind

    He earns double bachelor's degrees in computer science and economics at Peking University, co-founds a startup named GamesMind, and works as a research intern at Microsoft before joining Carnegie Mellon's master's program, class of 2021.

  2. 2020–2021

    Surveillance video at Carnegie Mellon

    As a Master of Language Technologies student advised by Alexander Hauptmann in CMU's School of Computer Science, he works on understanding surveillance videos in 3D. His team wins multiple professional challenges in 2020, including the CVPR AI City challenge and the TRECVID ActEV competition, and he is named to the Siebel Scholars class of 2021.

  3. 2022–2023

    The MAGVIT video transformer

    In December 2022, he posts MAGVIT, a Masked Generative Video Transformer that handles various video synthesis tasks with a single model, as first author. The work from Carnegie Mellon and Google Research is a CVPR 2023 Highlight.

  4. 2023–2024

    The MAGVIT-v2 tokenizer

    In October 2023, he is first author of "Language Model Beats Diffusion -- Tokenizer is Key to Visual Generation," which introduces MAGVIT-v2, a video tokenizer with a common token vocabulary for videos and images. With it, language models outperform diffusion models on ImageNet and Kinetics generation benchmarks, and the paper appears at ICLR in 2024.

  5. December 2023

    Co-authoring VideoPoet

    In December 2023, he co-authors VideoPoet, a decoder-only transformer language model that synthesizes video from conditioning signals such as images, videos, text and audio. Google Research introduces the model in a blog post that names him among its contributors.

  6. 2024

    A Ph.D. on video generation

    He completes his Ph.D. at Carnegie Mellon in 2024 with the thesis "Towards Multi-Task Multi-Modal Models: A Video Generative Perspective." The thesis reports the first instances of language models surpassing diffusion models in visual synthesis and a video tokenizer outperforming industry-standard codecs.

  7. July 2024

    A Best Paper at ICML

    In July 2024, VideoPoet receives a Best Paper award at ICML 2024. He is the second author, after Dan Kondratyuk.

  8. As of October 2026

    Staff Research Scientist on Gemini

    He is a Staff Research Scientist at Google DeepMind working on Gemini, with research interests in multimodal foundation models, especially video generation. He serves as an area chair at NeurIPS 2025, ICLR 2026, CVPR 2026 and ACL ARR 2026.

Career

  1. NowGoogle DeepMindStaff Research Scientist
  2. GamesMindCo-founder
  3. MicrosoftResearch Intern

Education

  1. 2024Carnegie Mellon UniversityPh.D.
  2. 2021Carnegie Mellon UniversityMaster of Language Technologies
  3. Peking UniversityBachelor's degree

From the biography.

  1. 16% of AI career moves in the 2020s went from a university to a companyIncludes Lijun Yu. We estimate that 16% of AI career moves in the 2020s went from a university to a company; 29% of moves starting at a university went to a company, and 76% ended the university role.
  2. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other universityIncludes Lijun Yu. The Massachusetts Institute of Technology trained or employed 44 of the 187 AI major-prize winners, more than any other university. We compare doctorate and 5+ year employment links, career timing and current workplaces for 187 major-prize and 1,111 best-paper winners.
  3. Google had the most people shaping AI every year from 2014 to 2025Includes Google. Our estimates of which organizations employed the most people shaping AI each year from 1955 to 2025, tracing the shift from university labs to companies.
  4. Google has the highest combined h-index among AI organizationsIncludes Google. We estimate that Google holds the highest combined h-index at 8,016, while universities account for 73% of the deduplicated total across AI organizations.

Sources and credits

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

  1. Scholar Profile – Page 1902 – Siebel Scholar Foundation siebelscholars.com. Checked 21 October 2026.
  2. [2212.05199] MAGVIT: Masked Generative Video Transformer arxiv.org. Checked 21 October 2026.
  3. MAGVIT magvit.cs.cmu.edu. Checked 21 October 2026.
  4. [2310.05737] Language Model Beats Diffusion -- Tokenizer is Key to Visual Generation arxiv.org. Checked 21 October 2026.
  5. Lijun Yu - Google Research research.google. Checked 21 October 2026.
  6. VideoPoet: A large language model for zero-shot video generation research.google. Checked 21 October 2026.
  7. [2312.14125] VideoPoet: A Large Language Model for Zero-Shot Video Generation arxiv.org. Checked 21 October 2026.
  8. Home - Lijun Yu 于力军 me.lj-y.com. Checked 21 October 2026.
  9. [2405.16728] Towards Multi-Task Multi-Modal Models: A Video Generative Perspective arxiv.org. Checked 21 October 2026.
  10. ICML 2024 Awards icml.cc. Checked 21 October 2026.