Anthropic says Claude is taking a growing role in developing the models that will succeed it.
As of August, Claude “leads” 26% of Anthropic’s model R&D work, up from virtually none earlier this year. Under the framework used by Anthropic, “leads” means Claude can complete most of a task end-to-end from a high-level prompt, with a human still supervising.
More than 90% of R&D work is now at least AI-collaborative.
That’s an important distinction: Anthropic says Claude is not yet fully autonomous in any measured area of its R&D. But the direction of travel is striking.
The company developed the new R&D Automation Index to track how much AI is contributing to building future AI, and potentially how close frontier labs are getting to recursive self-improvement.
Anthropic also acknowledges the methodology is still evolving, making comparisons across AI labs difficult for now.$英偉達(NVDA)$ $Meta Platforms, Inc.(META)$ $微軟(MSFT)$
Comments