Transformer pioneer Noam Shazeer bids farewell to Google once again to join OpenAI

According to Beating Monitoring, Google AI’s core figure and Gemini model technical director Noam Shazeer has left Google again and officially joined the competitor OpenAI. On Wednesday, OpenAI announced to employees that after Shazeer joins, he will focus on finding entirely new foundational architectures for large models and advancing the evolution of the Transformer architecture.

Shazeer is one of the co-authors of Google’s landmark 2017 paper “Attention Is All You Need.” The Transformer architecture proposed in the paper is the foundation of modern generative AI models such as ChatGPT, Claude, and Gemini. He joined Google in 2000 and was one of the earliest core engineers at the company. In 2021, due to security and brand considerations, Google refused to release Meena, a chatbot developed by him and Daniel De Freitas; Shazeer chose to leave and co-founded Character.AI.

In August 2024, to bring Shazeer back, Google reached a licensing and technology acquisition agreement with Character.AI worth as much as $2.7 billion. As a founder holding roughly 30% to 40% equity in Character.AI, Shazeer personally cashed out between $750 million and $1 billion. After returning to Google, he served as Vice President and co-technical director of Gemini, mainly responsible for model pre-training. However, less than two years after Google spent heavily to bring him back, Shazeer chose to leave again.

Shazeer’s decision to join is an important gain for OpenAI in the competition for talent, and it has also led to the loss of the key technical director of Google’s Gemini team. Large models are now approaching the limits of compute expansion for the Transformer architecture. OpenAI has put the original designer of the Transformer in charge of leading efforts to find alternatives, with the aim of gaining the upper hand in the next generation of large model architectures in order to respond to competition from Anthropic and Google.