Where Have the Eight Transformers' Pioneers Ended Up?

On June 18, Noam Shazeer announced his departure from Google to join OpenAI; two days later 2024 Nobel‑winning AlphaFold lead John Jumper left DeepMind for Anthropic. The news triggered a sharp market reaction, with Alphabet’s share price dropping more than 7% and analysts attributing the sell‑off to a talent exodus.

All eight co‑authors of the 2017 "Attention Is All You Need" paper have now left Google. A chart by X user Tyler Maran visualises their current affiliations, highlighting the dispersion of the team.

The authors’ contribution statements show distinct roles: Jakob Uszkoreit first proposed self‑attention; Ashish Vaswani and Illia Polosukhin designed and implemented the original Transformer; Noam Shazeer introduced scaled dot‑product attention, multi‑head attention and parameter‑free positional encodings; Niki Parmar built the early codebase and tensor2tensor framework; Llion Jones explored model variants and optimized inference; Łukasz Kaiser and Aidan Gomez rebuilt tensor2tensor, dramatically improving experimental results and research efficiency.

Career trajectories diverge dramatically: Vaswani moved from Google Brain to co‑found Adept AI (chief scientist), then Essential AI (CEO), and is rumored to be recruited by Nvidia for Nemotron work. Shazeer left Google, founded Character.AI, returned to DeepMind as engineering VP for Gemini, and later joined OpenAI for architecture research. Parmar left Google for Adept AI (CTO), co‑founded Essential AI, and in late 2024 joined Anthropic, contributing to Claude 3.7 Sonnet. Uszkoreit left Google to start Inceptive, an AI‑driven RNA‑design company now funded by Nvidia and AMD. Jones co‑founded Sakana AI in Tokyo, focusing on small collaborative models such as the Sakana Fugu. Gomez co‑founded Cohere, grew it into a multi‑billion‑dollar AI service provider, and now advocates for AI digital‑sovereignty and warns about misinformation amplification. Kaiser left Google for OpenAI, contributing to Codex, GPT‑4, and the o1‑o3 reasoning models, and argues that transformers can solve any problem given enough intermediate steps. Polosukhin left Google to launch NEAR Protocol, a blockchain platform, and now envisions AI agents as the future users of blockchain, calling for regulatory frameworks.

Recent rumors that Nvidia is quietly absorbing the core team of Essential AI, including Vaswani, underscore the fierce competition for AI talent. Across interviews and public statements, the authors repeatedly note that no one truly believes the Transformer is the final architecture; they continue searching for the next breakthrough, a sentiment that forms the lasting legacy of their seminal paper.