Jan Leike joined Anthropic as an alignment team lead
Following his resignation from OpenAI, Jan Leike recently announced he will join Anthropic to lead a team resembling OpenAI's now-defunct Superalignment team. According to Leike, the team will work on scalable oversight, weak-to-strong generalization, and automated alignment research.
Amid the controversy surrounding OpenAI's new Safety and Security Committee, Jan Leike announced on the social network X that he has joined Anthropic, where he will lead a team working on scalable oversight, weak-to-strong generalization, and automated alignment research. Leike also leveraged his announcement to let the community know he was looking for talent to hire. According to a report, Leike will report to Anthropic’s chief science officer, Jared Kaplan, and researchers currently working in scalable oversight will be expected to report to Leike as his team shapes up.
The team sounds much like OpenAI's former Superalignment team, which seems to have been slowly disbanded. The fate of OpenAI's Superalignment team surely influenced Leike's decision to resign, as can be seen from a series of X posts where he shared more on his disagreements with the company on the company's priorities. Among other things, Leike describes the work with his team during the last months as "sailing against the wind" as they sometimes struggled to get the computing resources needed to perform their research.
Tellingly, Anthropic co-founder and CEO Dario Amodei left his position as VP of research at OpenAI following a disagreement on OpenAI's increasing focus on maximizing profit. Amodei went on to found Anthropic, a company that prides itself on being more security-focused than OpenAI, with Daniela Amodei and a small group of former OpenAI employees whose departure followed his.
