Jason Wei, the researcher best known as first author of the 2022 paper "Chain-of-Thought Prompting Elicits Reasoning in Large Language Models," is leaving Meta to join Mercor as chief scientist. Mercor co-founder and CEO Brendan Foody announced the hire on X on Tuesday, and Wei confirmed the move in his own post shortly after.
Wei said he had spent "a great year-plus" at Meta's TBD Lab, one of the four teams inside Meta Superintelligence Labs working on the company's frontier models. In explaining the move, he pointed at the data business rather than at another frontier lab: as more companies begin training their own AI models, he argued, data will shape the next frontier of AI development.
The trajectory is notable for where it does not go. Wei has moved from Google Brain, where the chain-of-thought paper was published, to OpenAI, and then to Meta's superintelligence organization. His new employer is not Google DeepMind, OpenAI or another model lab, but a three-year-old startup built around expert data and AI-enabled recruiting.
That startup is not small. Mercor raised a $750 million Series D at a $20 billion valuation just last week, with participation from Nvidia and Apollo, less than a year after doubling its valuation from $2 billion to $10 billion. The company matches experts with AI labs that need domain-specific training data, and it has mapped thousands of AI benchmarks onto real American job tasks to price that expertise.
The hire reads as a statement about where Mercor wants to go. If Wei's thesis holds — that model quality increasingly depends on who can source, structure and verify expert data at scale — then putting a chain-of-thought author at the head of research gives Mercor a credible claim to be more than a data vendor. It also signals that frontier-lab alumni now see data companies as research institutions in their own right, a shift that would have been hard to imagine two years ago.
For Meta, the departure is another loss from a lab that has already watched several senior researchers leave this year. For the wider market, it is the clearest signal yet that the competition for AI talent has moved beyond the half-dozen frontier labs — and that the people who wrote the field's most influential papers now believe the next edge is in the data pipeline, not the model architecture.
Comments (0)
Log in to join the discussion
Log InNo comments yet