Jan Leike, a leading AI scientist that previously this month surrendered from OpenAI prior to openly slamming the firm’s strategy to AI safety and security, has actually signed up with OpenAI competing Anthropic to lead a brand-new “superalignment” group.
In a message on X, Leike stated that his group at Anthropic will certainly concentrate on different facets of AI safety and security and protection, especially “scalable oversight,” “weak-to-strong generalization” and automated placement research study.
A resource accustomed to the issue informs TechCrunch that Leike will certainly report straight to Jared Kaplan, Anthropic’s primary scientific research police officer, which Anthropic scientists presently working with scalable oversight– methods to regulate massive AI’s habits in foreseeable and preferable methods– will certainly relocate to report to Leike as Leike’s group rotates up.
In numerous methods, Leike’s group appears comparable in goal to OpenAI’s recently-dissolved Superalignment group. The Superalignment group, which Leike co-led, had the enthusiastic objective of addressing the core technological obstacles of managing superintelligent AI in the following 4 years, however typically discovered itself hamstringed muscle by OpenAI’s management.
Anthropic has actually typically tried to place itself as even more safety-focused than OpenAI.
Anthropic’s chief executive officer, Dario Amodei, was when the VP of research study at OpenAI, and apparently divided with OpenAI after a difference over the firm’s instructions– particularly OpenAI’s expanding industrial emphasis. Amodei brought with him a variety of ex-OpenAI workers to release Anthropic, consisting of OpenAI’s previous plan lead Jack Clark.
