PeopleofSuperintelligence

Researcher

Zihang Dai

Zihang Dai is an AI researcher whose work spans semi-supervised learning, language models and computer vision. He coauthored Unsupervised Data Augmentation, Transformer-XL and XLNet in 2019, then coauthored Funnel-Transformer, CoAtNet and the FLASH design for long sequences. At Carnegie Mellon, he earned a Ph.D. in Language and Information Technology in 2020 and moved to Google as a researcher. He joined xAI's founding team in July 2023, remained an xAI founding member in February 2025, and left the company in March 2026.

Badges

Timeline

  1. 2019

    Unsupervised data augmentation

    In 2019, he coauthors Unsupervised Data Augmentation for Consistency Training, which uses RandAugment and back-translation for consistency training.

  2. 2019

    Transformer-XL extends context

    In 2019, he coauthors Transformer-XL, a language-model architecture that learns dependencies beyond fixed-length context through segment-level recurrence and a new positional encoding.

  3. 2019

    XLNet learns bidirectionally

    In 2019, he coauthors XLNet, a generalized autoregressive pretraining method that learns bidirectional contexts by maximizing the expected likelihood over all permutations of the factorization order.

  4. 2020

    A Carnegie Mellon PhD

    In 2020, he earns a Ph.D. in Language and Information Technology at Carnegie Mellon's Language Technologies Institute, with Yiming Yang as his advisor. After graduating in 2020, he joins Google as a researcher.

  5. 2020

    Funnel-Transformer compresses context

    In 2020, he coauthors Funnel-Transformer, which compresses the sequence of hidden states to reduce computation.

  6. 2021

    CoAtNet for computer vision

    In 2021, he coauthors CoAtNet, a family of vision models that combines depthwise convolution with self-attention.

  7. 2022

    Transformer quality in linear time

    In 2022, he coauthors Transformer Quality in Linear Time, which proposes a gated attention unit and a linear approximation for long sequences. Its FLASH model matches improved Transformers on short and long contexts.

  8. July 2023–March 2026

    xAI founding team and exit

    In July 2023, he joins xAI's 12-member founding team as a former Google research scientist. In February 2025, he is an xAI founding member. In March 2026, he leaves the company during its personnel overhaul.

Career

  1. 2023–2026xAIFounder
  2. 2020–2023GoogleResearch Scientist

Education

  1. 2020Carnegie Mellon UniversityPh.D.

From the biography.

  1. Google alumni have founded or led 133 other organizations, more than any other AI labIncludes Zihang Dai. We estimate that alumni of 17 major AI labs went on to found or lead 350 other organizations, including 226 they founded, and we trace 472 career moves directly between 23 labs since 2010.
  2. Of 943 people shaping AI who worked at Google, 446 work there nowIncludes Zihang Dai. Our estimates of where the 943 people shaping AI who worked at Google and Google DeepMind work now: 446 at Google and 452 across 278 other organizations.
  3. Companies supply the most prior-work hires at 7 of the 9 frontier AI labsIncludes Zihang Dai. Our estimates of the organizations where 550 people worked directly before joining 9 frontier AI labs across 620 career transitions, comparing the main feeder organizations per lab.
  4. 11% of AI doctorate holders from 2010 to 2016 founded or led a company or lab by 10 years after graduationIncludes Zihang Dai. We estimate how often AI doctorate holders founded or led a company or lab by 10 years after graduation, how waiting times compare between founders and top jobs, and which universities trained the most.

Colleagues

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

Sources and credits

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

  1. Unsupervised Data Augmentation for Consistency Training arxiv.org. Checked 21 October 2026.
  2. Transformer-XL: Attentive Language Models beyond a Fixed-Length Context - ACL Anthology aclanthology.org. Checked 21 October 2026.
  3. XLNet: Generalized Autoregressive Pretraining for Language Understanding proceedings.neurips.cc. Checked 21 October 2026.
  4. Zihang Dai - Language Technologies Institute - School of Computer Science - Carnegie Mellon University lti.cs.cmu.edu. Checked 21 October 2026.
  5. Funnel-Transformer: Filtering out Sequential Redundancy for Efficient Language Processing arxiv.org. Checked 21 October 2026.
  6. CoAtNet: Marrying Convolution and Attention for All Data Sizes arxiv.org. Checked 21 October 2026.
  7. Transformer Quality in Linear Time proceedings.mlr.press. Checked 21 October 2026.
  8. Elon Musk Announces xAI: Who's On the 12-Man Founding Team? observer.com. Checked 21 October 2026.
  9. Which Chinese AI experts are hidden behind Elon Musk's Grok3? eu.36kr.com. Checked 21 October 2026.
  10. 'Not built right the first time' — Musk's xAI is starting over again, again techcrunch.com. Checked 21 October 2026.