Anthropic Recruits Former OpenAI Safety Lead for New "Superalignment" Team
Anthropic Recruits Former OpenAI Safety Lead for New "Superalignment" Team
Anthropic has made a significant move in the AI industry by hiring Jan Leike, a former safety lead at OpenAI, to head a newly formed "superalignment" team. Leike, who recently resigned from OpenAI, had publicly criticized the company's approach to AI safety, citing concerns over its methods and direction. At Anthropic, he will focus on critical aspects of AI safety, including scalable oversight, weak-to-strong generalization, and automated alignment research.
Leike's team at Anthropic is set to concentrate on ensuring that large-scale AI behaves predictably and aligns with desired outcomes. This involves developing robust techniques for scalable oversight and pushing forward research in automated alignment. These efforts are aimed at making AI systems more reliable and secure, a mission that closely mirrors the objectives of the now-dissolved Superalignment team at OpenAI, which Leike co-led.
Reporting directly to Anthropic’s Chief Science Officer, Jared Kaplan, Leike's appointment highlights Anthropic's commitment to prioritizing AI safety over commercial interests. This focus is a notable contrast to OpenAI's increasingly commercial orientation, a point of contention that previously led Dario Amodei, Anthropic's CEO and former VP of Research at OpenAI, to part ways with the organization. Amodei's departure brought several ex-OpenAI employees to Anthropic, including policy lead Jack Clark, further emphasizing the company's safety-first approach.
Anthropic’s strategic direction under Leike's leadership is expected to advance the field of AI safety significantly. By consolidating efforts on scalable oversight and alignment, Anthropic aims to tackle the pressing challenges of managing superintelligent AI systems. This move underscores the competitive landscape in AI research, with companies like Anthropic and OpenAI vying to set the standard for safe and effective AI development.