Noam Shazeer, a key figure in Google AI and co-technical lead of the Gemini model, has left Google once again to join rival OpenAI. OpenAI announced to employees on Wednesday that Shazeer will focus on finding entirely new foundational architectures for large models and pushing the evolution of the Transformer.
https://twitter.com/NoamShazeer/status/2067400851438932297
Shazeer was a co-author of Google's landmark 2017 paper "Attention Is All You Need," which introduced the Transformer architecture that powers modern generative AI models like ChatGPT, Claude, and Gemini. He joined Google in 2000 and was one of its earliest core engineers. In 2021, after Google refused to release the Meena chatbot he developed with Daniel De Freitas over safety and brand concerns, Shazeer left to co-found Character.AI.
In August 2024, Google inked a massive $2.7 billion licensing and technology acquisition deal with Character.AI to bring Shazeer back. As a founder holding roughly 30% to 40% equity in Character.AI, he personally cashed out between $750 million and $1 billion. He returned to Google as a vice president and co-technical lead of Gemini, overseeing model pre-training. But less than two years after Google spent a fortune to reel him in, Shazeer is heading for the exit again.
Shazeer's defection is a major talent win for OpenAI and a significant loss for Google's Gemini team. With large models hitting the compute scaling limits of the Transformer architecture, OpenAI is betting that having the original Transformer designer lead the search for alternatives will give it a head start in the next-generation model race against Anthropic and Google.