AI People Wiki

Builders & open source

Barret Zoph

First author of neural architecture search at Google Brain; earlier VP of Research (Post-Training) at OpenAI

Barret Zoph's homepage describes earlier roles as VP of Research (Post-Training) at OpenAI, where Zoph's team worked on alignment, tool use, evaluations, and the models shipped in ChatGPT and the API, and before that as a Staff Research Scientist at Google Brain working on large sparse language models and AutoML. With Quoc V. Le at Google Brain, Zoph wrote Neural Architecture Search with Reinforcement Learning (2016), and Zoph's homepage lists Switch Transformers and ST-MoE among Zoph's sparse-model papers. OpenAI's GPT-4o and o1 system cards credit Zoph as a post-training organization lead and in executive leadership, respectively.1234

At a glance

Current role
None confirmed in the sources
Education
USC
Work listed
2
Sources cited
9
Profile reviewed

Career

Before joining Google Brain, Zoph worked at the Information Sciences Institute with Kevin Knight and Daniel Marcu on statistical machine translation. A 2016 paper with Deniz Yuret, Jonathan May and Knight presented a transfer learning method for neural machine translation of low-resource languages.15

At Google Brain, Zoph was a Staff Research Scientist who trained large sparse language models and worked on AutoML in both research and product; Zoph's homepage also mentions serving as a research technical lead for MuM. Zoph later became VP of Research (Post-Training) at OpenAI, where the team worked on alignment, tool use, evaluations, search and multimodality and trained the models released in ChatGPT and the API. The GPT-4o and o1 system cards list Zoph in post-training leadership and executive leadership roles.134

Research

Zoph's work on neural architecture search began with a 2016 paper with Quoc V. Le that used a recurrent network trained with reinforcement learning to design neural network architectures. A 2017 follow-up learned architectures on a small dataset and transferred them to large-scale image recognition, and the 2018 AutoAugment paper applied a similar search to data augmentation policies.267

Zoph also worked on sparse mixture-of-experts language models. Zoph co-authored "Switch Transformers" (2021) with William Fedus and Noam Shazeer, which simplified expert routing to scale models to very large parameter counts, and was co-first author of "ST-MoE" (2022), which addressed training instability and fine-tuning issues in sparse expert models.891

Work and contributions

  • Neural architecture search Published research

    First author, with Quoc V. Le at Google Brain, of the 2016 paper that trains a recurrent network with reinforcement learning to generate neural network architectures; Zoph's homepage says this work was an impetus for Google's AutoML.21

  • OpenAI post-training (GPT-4o and o1) Proprietary product

    Credited as a post-training organization lead in the GPT-4o System Card (2024) and in executive leadership in the OpenAI o1 System Card (2024).34

Timeline

  1. Posted "Transfer Learning for Low-Resource Neural Machine Translation" as first author.5
  2. Posted "Neural Architecture Search with Reinforcement Learning" with Quoc V. Le.2
  3. Posted "Learning Transferable Architectures for Scalable Image Recognition" as first author.6
  4. Co-authored AutoAugment, a method for learning data augmentation policies from data.7
  5. Co-authored "Switch Transformers", a paper on scaling sparse mixture-of-experts models.8
  6. Posted "ST-MoE" on stable and transferable sparse expert models as co-first author.9
  7. Credited in post-training leadership in the GPT-4o System Card.3
  8. Credited in executive leadership in the OpenAI o1 System Card.4

Documented roles

The sources do not confirm a current role; these are the roles they document.

Companies in their history

Suggest a correction or removal

Profiles cover public professional work only, and each statement cites a source. If something is out of date, missing or wrong, tell us and link a source. People can also ask for their profile to be corrected or removed.

References

9 sources · profile reviewed Oct 21, 2026 · data updated Oct 21, 2026

  1. Barret Zoph — homepageBarret Zoph · Official page · accessed Oct 21, 2026
  2. Neural Architecture Search with Reinforcement Learning (arXiv 1611.01578)arXiv · Paper · published Nov 5, 2016 · accessed Oct 21, 2026
  3. GPT-4o System Card (arXiv 2410.21276), contributionsarXiv · Paper · published Oct 25, 2024 · accessed Oct 21, 2026
  4. OpenAI o1 System Card (arXiv 2412.16720), authorship and credit attributionarXiv · Paper · published Dec 21, 2024 · accessed Oct 21, 2026
  5. Transfer Learning for Low-Resource Neural Machine Translation (arXiv 1604.02201)arXiv · Paper · published Apr 8, 2016 · accessed Oct 21, 2026
  6. Learning Transferable Architectures for Scalable Image Recognition (arXiv 1707.07012)arXiv · Paper · published Jul 21, 2017 · accessed Oct 21, 2026
  7. AutoAugment: Learning Augmentation Policies from Data (arXiv 1805.09501)arXiv · Paper · published May 24, 2018 · accessed Oct 21, 2026
  8. Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity (arXiv 2101.03961)arXiv · Paper · published Jan 11, 2021 · accessed Oct 21, 2026
  9. ST-MoE: Designing Stable and Transferable Sparse Expert Models (arXiv 2202.08906)arXiv · Paper · published Feb 17, 2022 · accessed Oct 21, 2026