Report says Anthropic is using AI to speed up its own AI research

In short: A new report says Anthropic is using its own AI systems to help build better AI, and the pace of improvement is increasing.

What's going on

AI safety group METR published a report after spending 11 weeks looking inside Anthropic, the company behind the Claude chatbot. The report focuses on how Anthropic uses AI tools to help its researchers do their work.

METR says Anthropic’s AI models are improving about twice as quickly as they were a year ago. Based on benchmark scores (standard tests that compare models, like the same exam given each year), METR suggests Anthropic made about two years of past progress in just the last year.

Researchers say this is some of the clearest evidence yet of AI speeding up AI development. You can think of it like using a power tool to build an even better power tool, then using the new one to build the next version faster.

The report also notes limits on what the public can see. METR allowed Anthropic to remove some parts of the final write-up, and the article says Anthropic used that option often. That raises questions about how much outside observers can truly know about the speed of development and the possible risks.

What to watch

Anthropic’s CEO, Dario Amodei, has publicly called for slowing down AI development, even as the company’s internal work appears to be accelerating. Watch for whether more labs allow independent reviewers inside, and whether future reports include fewer redactions so policymakers and the public can better judge what is happening.

Source: The Verge AI