The co-author of the paper that made current-generation AI possible is now at
OpenAI.
Shazeer announced the move directly on his X account on June 18. Sam Altman welcomed him the
same day with a post of his
own, saying he’d wanted to work with Shazeer since the beginning of OpenAI. The announcement came less than two years after Google paid approximately $2.7
billion, for a combination of Shazeer’s return and non-exclusive rights to
Character.AI’s technology, according to Business Insider and Wall Street
Journal reporting from August 2024. Shazeer’s specific role and title at
OpenAI have not been announced. Don’t expect that detail soon, OpenAI rarely
discloses research appointments before the work is ready.
For context on why the AI industry is paying attention to a personnel announcement:
Shazeer is a co-author of “Attention Is
All You Need” (Vaswani et al., 2017), the paper that introduced the transformer
architecture. Every major frontier model in production today, GPT, Claude, Gemini,
Llama, runs on transformer architecture. His contributions to mixture-of-experts
architectures and fast attention mechanisms extend the practical influence further. This isn’t a symbolic hire. It’s a research talent move with direct implications for
model development trajectories.
Shazeer Departure: Stakeholder Positions
The catch is that “implications” is about as specific as we can get right now. Shazeer’s role, research focus, and team placement at OpenAI are unconfirmed. Any
analysis of what he’ll work on at OpenAI is speculation on top of an announcement.
What’s confirmed: Google loses a researcher it valued at $2.7 billion by its own
revealed preference. OpenAI gains him. Altman’s framing, that he’d wanted to work
with Shazeer since OpenAI’s founding, suggests this wasn’t opportunistic recruiting
on OpenAI’s part. Whether Google’s Gemini research leadership experiences practical
disruption depends on how central Shazeer was to active model development, and that
isn’t disclosed.
What to Watch
What to watch. Two signals matter. First, any formal role announcement from
OpenAI, the specific research area will tell you more about where the company
thinks its frontier work needs strengthening. Second, whether Google responds with
a counter-hire or internal promotion to its model research leadership. Executive and
research departures at this level tend to trigger visible organizational responses.
TJS synthesis
The transformer co-author is now at the company building on
transformer architecture at scale. That’s a concrete competitive shift, not just a
headline. Enterprise teams on Google Gemini roadmaps should watch OpenAI’s next
model announcements, not because Shazeer’s contributions will show up
immediately, but because research talent of this depth tends to shape architecture
decisions 12-to-24 months out. Wait for the role announcement before drawing
conclusions about specific model capability trajectories, but put the date of hire
in your competitive intelligence log.