Jan Leike, a number one AI researcher who earlier this month resigned from OpenAI earlier than publicly criticizing the corporate’s method to AI security, has joined OpenAI rival Anthropic to steer a brand new “superalignment” crew.
In a put up on X, Leike stated that his crew at Anthropic will concentrate on varied points of AI security and safety, particularly “scalable oversight,” “weak-to-strong generalization” and automatic alignment analysis.
A supply conversant in the matter tells TechCrunch that Leike will report on to Jared Kaplan, Anthropic’s chief science officer, and that Anthropic researchers at the moment engaged on scalable oversight — methods to regulate large-scale AI’s conduct in predictable and fascinating methods — will transfer to report back to Leike as Leike’s crew spins up.
In some ways, Leike’s crew sounds comparable in mission to OpenAI’s recently-dissolved Superalignment crew. The Superalignment crew, which Leike co-led, had the bold objective of fixing the core technical challenges of controlling superintelligent AI within the subsequent 4 years, however usually discovered itself hamstrung by OpenAI’s management.
Anthropic has usually tried to place itself as extra safety-focused than OpenAI.
Anthropic’s CEO, Dario Amodei, was as soon as the VP of analysis at OpenAI, and reportedly cut up with OpenAI after a disagreement over the corporate’s route — particularly OpenAI’s rising industrial focus. Amodei introduced with him quite a few ex-OpenAI workers to launch Anthropic, together with OpenAI’s former coverage lead Jack Clark.