PeopleofSuperintelligence

Lab leader

Andrew M. Dai

Also known as Andrew Dai

Andrew Dai is an AI researcher and startup leader whose work spans recurrent-network pretraining, language models, and Gemini's data program. He and Quoc Le published a recurrent-network pretraining method in 2015. At Google, he contributed to Gmail's Smart Compose, co-led GLaM and PaLM 2 pretraining, and led Gemini's data area. As of October 2026, he is co-founder and CEO of Elorian AI, where his team builds multimodal models for visual reasoning.

Now at
Elorian AI
Works
44

Publication numbers from OpenAlex, as of 15 October 2026. Citations and h-index are hidden because this OpenAlex record may be incomplete or mix in other authors.

Timeline

  1. 2015

    Pretraining recurrent networks

    He and Quoc V. Le publish Semi-supervised Sequence Learning in 2015, using unlabeled data to pretrain recurrent networks before supervised training.

  2. 2018

    Smart Compose in Gmail

    In 2018, he contributes to Smart Compose, a Gmail feature that uses machine learning to offer sentence completions as people type.

  3. 2021

    Scaling with GLaM

    In 2021, he co-leads GLaM, a sparsely activated mixture-of-experts language model with 1.2 trillion parameters.

  4. 2023

    PaLM 2 pretraining

    In 2023, he co-leads pretraining for PaLM 2, a language model with multilingual and reasoning capabilities.

  5. 2023

    Leading Gemini data

    By 2023, he leads Gemini's data area for Google's multimodal model family, which supports image, audio, video, and text understanding.

  6. As of October 2026

    Building Elorian AI

    He is Elorian AI's co-founder and CEO, where the team builds specialized multimodal foundation models for visual reasoning. In 2026, after 12 years at Google Brain and DeepMind, he leaves Google DeepMind to build the startup.

Sources and credits

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

  1. Semi-supervised Sequence Learning arxiv.org. Checked 15 October 2026.
  2. Smart Compose: Using Neural Networks to Help Write Emails research.google. Checked 15 October 2026.
  3. GLaM: Efficient Scaling of Language Models with Mixture-of-Experts arxiv.org. Checked 15 October 2026.
  4. Team — Elorian elorian.ai. Checked 15 October 2026.
  5. PaLM 2 Technical Report arxiv.org. Checked 15 October 2026.
  6. Gemini: A Family of Highly Capable Multimodal Models arxiv.org. Checked 15 October 2026.
  7. Reasoning Doesn’t Start With Language thedataexchange.media. Checked 15 October 2026.
  8. The Lab Builders Are Leaving: Andrew Dai & Elorian AI opensourceceo.com. Checked 15 October 2026.
  9. How a former DeepMind researcher raised at a $300M pre-seed valuation before launching a product techcrunch.com. Checked 15 October 2026.