Wei was involved in OpenAI’s deep and o1 research models. He joined OpenAI 2023, after a stint with Google, where he worked in chain-of-thoughtresearch, which involves teaching an AI model how to process complex queries one step at a time. Wei, who is a self-described “diehard” for reinforcement learning at OpenAI, became a proponent of this method, which involves training or refining AI models with positive or negative feedback. It’s a promising field of AI research, and one that several of the researchers Meta hired for its superintelligence group specialize in.
A source told WIRED that Hyung Won Chung will also join Meta. Multiple sources confirm both Wei and Chung’s OpenAI Slack profiles have been deactivated. WIRED’s requests for comment were not immediately answered by OpenAI, Meta Wei and Chung. Chung, according to Chung’s personal website, worked on the same projects as Wei at OpenAI, including deep research, OpenAI’s o1 Model, and OpenAI’s o1 model. According to the website, his research focuses primarily on reasoning and agents. Chung and Wei worked together at Google, and both joined OpenAI around the same time, according to their LinkedIn profiles.
According to multiple sources, WIRED has confirmed that Wei and Chung share a close relationship. Meta has poached researchers who have worked together in the past for its superintelligence lab. This includes a trio of OpenAI researchers from Switzerland, who joined ChatGPT from Google.
Meta is on a poaching rampage, offering top AI talent up to $300 million in four years. WIRED reported that Meta CEO Mark Zuckerberg had sent an internal memo late last month to staff outlining a new plan for the company’s AI efforts. The memo included a list with new staffers, many of whom were recruited from OpenAI.
OpenAI is fighting back against the hiring frenzy, which shows no signs of slowing. WIRED reported just last week that OpenAI had hired four high-ranking engineering staff from Tesla, xAI and Meta.
Wei shared on Tuesday a social media post reflecting on “an important lesson” reinforcement learning taught him about “how to live my life.”
Wei wrote that imitation is good in life (and when building AI model), and you must do it at first. But “beating your teacher requires walking your path and taking risks and rewarding the environment.”
