Noam Shazeer
Researcher, OpenAI
Co-wrote the transformer paper, co-founded Character.AI, co-led Gemini at Google, then joined OpenAI in 2026.
Noam Shazeer studied mathematics and computer science at Duke and joined Google in 2000 as one of its early engineers, where his first work improved the search engine's spelling correction. At Google Brain he worked on making neural networks larger and cheaper to run. In January 2017 he was lead author of the paper on sparsely gated mixture-of-experts layers, which routed each input through a few of thousands of sub-networks, and that June he was one of the eight equal contributors to "Attention Is All You Need," the transformer paper. In November 2019 he wrote, as sole author, the paper introducing multi-query attention, which made transformer text generation faster.
With Daniel De Freitas he built Meena, a conversational chatbot. The Wall Street Journal reported that Google declined to release it publicly over safety and fairness concerns, and that Shazeer had predicted it could one day replace search. The pair left in 2021 and founded Character.AI, where users chat with AI personas; it reached a valuation of one billion dollars by 2023.
In August 2024 Google signed a non-exclusive license to Character.AI's models and hired Shazeer, De Freitas and part of the research team into Google DeepMind. Google described it as welcoming back "a preeminent researcher"; the Journal reported that Google paid about 2.7 billion dollars, in large part to bring him back. In May 2025 Bloomberg reported that the Justice Department was examining whether the arrangement was structured to avoid merger review; Google said Character.AI remained "a separate company." Shazeer became a Gemini co-lead. Families who sued Character.AI over harms to teenagers named him, De Freitas and Google as defendants; the parties agreed to settle in January 2026, on undisclosed terms.
On 18 June 2026 he announced he was leaving Google for OpenAI, calling it "a difficult decision to move on." OpenAI had not disclosed his title by September 2026.
Known for
Sparsely gated mixture of experts
A January 2017 paper showing that a network with up to 137 billion parameters could stay cheap to run by activating only a few expert sub-networks per input, a design later used in many frontier models.
The transformer
One of eight equal contributors to the June 2017 paper that replaced recurrence with attention; the architecture underlies GPT, Gemini and most large language models.
Meena and Character.AI
Built the chatbot Google withheld, then co-founded a company that put chatbot personas in front of millions of users.
On the record
Shazeer said in February 2025 that people should be free to do as much as possible with AI systems, but not to take a model and turn it into mass-scale harm.
“if you let somebody take your thing and create a million evil software engineers, then that doesn't empower people because they're going to hurt others with a million evil software engineers. So I'm against that.”
Dwarkesh Podcast, with Jeff Dean, February 2025 (01:37), 2025
Career
-
2000-2021
Google
Engineer, later at Google Brain
-
2021-2024
Character.AI
Co-founder and Chief Executive
-
2024-2026
Vice President of Engineering and Gemini co-lead
-
2026-present
Researcher (title not disclosed)
Sources
- Jeff Dean and Noam Shazeer, 25 years at Google (Dwarkesh Podcast, February 2025)
- Outrageously large neural networks, the sparsely-gated mixture-of-experts layer (arXiv, January 2017)
- Attention is all you need (arXiv, June 2017)
- Ex-Google engineers who founded Character.AI rejoin company with new AI partnership (CNBC, August 2024)
- Google paid 2.7 billion dollars to rehire AI genius who left after company refused to release his bot (Yahoo Finance, summarising the Wall Street Journal, September 2024)
- US DOJ probing Google over Character.AI deal (Yahoo Finance, May 2025)
- Character.AI and Google agree to settle lawsuits over teen mental health harms (CNN, January 2026)
- Google Gemini co-lead Noam Shazeer leaves for OpenAI (CNBC, June 2026)
- Fast transformer decoding, one write-head is all you need (arXiv, November 2019)
Brief profiles cover a person's professional record: roles, work and public statements, each from a source listed here. Full profiles, with positions across the debates, are kept for the people the site follows most closely.