OpenAI researcher Jason Wei is becoming a member of Meta’s new superintelligence lab, in keeping with a number of sources accustomed to the matter.
Wei labored on OpenAI’s o3 and deep analysis fashions, in keeping with his private web site. He joined OpenAI in 2023 after a stint at Google, the place he labored on chain-of-thought analysis, which entails coaching an AI mannequin to course of advanced queries step-by-step. At OpenAI, Wei grew to become a self-described “diehard” for reinforcement studying, a technique of coaching or refining an AI mannequin with constructive or unfavorable suggestions. It’s change into a promising space of AI analysis—one which a number of of the researchers Meta has employed for its superintelligence workforce specialise in.
One supply tells WIRED that one other OpenAI researcher, Hyung Gained Chung, will even be becoming a member of Meta. A number of sources verify that each Wei and Chung’s inside OpenAI Slack profiles are at the moment deactivated. OpenAI, Meta, Wei, and Chung didn’t instantly reply to requests for remark from WIRED.
Chung labored on among the identical initiatives at OpenAI as Wei, together with deep analysis and OpenAI’s o1 mannequin, in keeping with Chung’s private web site. His analysis is primarily targeted on reasoning and brokers, the web site says. Chung overlapped with Wei at Google as properly, and joined OpenAI similtaneously Wei, per their LinkedIn profiles.
A number of sources inform WIRED that Wei and Chung have an in depth working relationship. Meta has beforehand poached teams of researchers which have expertise working collectively for its new superintelligence lab, together with a trio from OpenAI’s Switzerland workplace that joined the ChatGPT maker from Google.
Meta has been happening a poaching spree over the previous month, providing as much as $300 million over 4 years to high AI expertise. WIRED reported late final month that Meta CEO Mark Zuckerberg despatched an inside memo to employees that laid out a recent plan for the corporate’s AI efforts. It included an inventory of latest staffers for the superintelligence workforce, most of whom had been recruited from OpenAI.
The hiring frenzy exhibits no indicators of slowing down, and OpenAI has been preventing again. Simply final week, WIRED reported that OpenAI had recruited 4 high-ranking engineers from Tesla, xAI, and Meta.
On Tuesday, Wei shared a publish on social media reflecting on what he known as “an essential lesson” that reinforcement studying taught him “about tips on how to dwell my very own life.”
In life, (and when constructing AI fashions), imitation is sweet and it’s important to do it at first, Wei wrote. However “beating the trainer requires strolling your personal path and taking dangers and rewards from the surroundings.”
Keep forward of the curve with NextBusiness 24. Discover extra tales, subscribe to our e-newsletter, and be part of our rising neighborhood at nextbusiness24.com

