Source: Jacob Devlin, who left Google to join OpenAI in January 2023 after complaining internally Bard was being trained on ChatGPT data, has returned to Google
Jon Victor / The Information :
Context & Ripple Effects
Jacob Devlin's return closes a loop that opened with his January 2023 exit, when he reportedly complained internally that Bard was being trained on ChatGPT data before decamping to OpenAI. His departure landed mid-crisis: Google had just pulled executives into emergency sessions with Larry Page and Sergey Brin and forced Brain and DeepMind into the joint Gemini project after Bard's stumble.
The return also lands amid a two-way talent current: OpenAI has been aggressively poaching Google employees for its search-with-citations team, while its own hires keep churning — Shivakumar Venkataraman left within a year of joining, and John Schulman has now exited Anthropic too.
First-order effects
- Google regains a researcher who knows both its Bard-era practices and OpenAI's internals, while OpenAI loses a hire it won at the height of its recruiting push against Google.
Second-order effects
- OpenAI's poaching campaign against Google now runs against evidence that its own seats don't hold, which weakens the pitch it uses on Google employees weighing a jump.
Third-order effects
- If boomerang moves like this become routine, frontier-lab careers normalize as circulation between rivals rather than one-way defection — and each return carries competitive intelligence across the aisle, raising the stakes of how labs handle internal complaints about training data.
The trend: Frontier AI talent is shifting from a one-way drain out of Google toward bidirectional churn among rival labs, with returns becoming as common as defections.