Noam Shazeer

HE HELPED INVENT THE TRANSFORMER, THEN DEFECTED TO OPENAI

Shazeer spent decades inside Google's AI machinery, co-authored the Transformer, left to build Character.AI, came back to co-lead Gemini — then jumped to OpenAI in 2026, carrying one of the architecture's inventors across the frontier-model war. His other hit: mixture-of-experts — the trick of waking only the network's needed experts per example, which became the quiet scaling engine behind the biggest models.

1. 2000

   HE JOINS GOOGLE BEFORE MODERN AI EXISTS

   Shazeer joined Google in 2000, working on systems including the company's spelling corrector and advertising technology before the deep-learning era.

   SOURCES: Noam Shazeerfirst-party

2. 2016

   HE MAKES GIANT MODELS SPARSE

   Shazeer developed the sparsely gated mixture-of-experts approach from his 2017 paper 'Outrageously Large Neural Networks,' activating only parts of a huge neural network for each example, a design that later became central to many large models.

   SOURCES: Noam Shazeerfirst-party

3. 2017

   HE CO-AUTHORS THE TRANSFORMER

   Shazeer was one of the eight authors of 'Attention Is All You Need,' contributing to the architecture that became the foundation of the modern language-model boom.

   SOURCES: Google Researchfirst-party · Noam Shazeerfirst-party

4. Aug 2024

   GOOGLE BRINGS HIM BACK TO HELP RUN GEMINI

   After Character.AI struck a licensing deal with Google and Shazeer returned, Google made him a technical co-lead of its Gemini model effort.

   SOURCES: Reutersreported

5. Jun 18, 2026

   THEN HE JUMPS FROM GEMINI TO OPENAI

   Shazeer left Google and its Gemini leadership team to join OpenAI, moving one of the Transformer authors directly across the frontier-model rivalry.

   SOURCES: Reutersreported