Anthropic AI development doubles speed despite CEO safety warnings

The rapid acceleration in Anthropic's AI development, detailed by the METR evaluation group, centers on a process called recursive self-improvement. This mechanism involves Anthropic's existing Claude chatbot actively participating in the creation of subsequent AI models. Effectively, the company is leveraging its current AI capabilities to enhance and build more advanced versions of its own technology, directly contributing to the observed doubling of development speed.

This method of AI-assisted development has implications for the pace of technological progress. METR, which gained access to Anthropic's internal data and staff, warned that unchecked self-improving AI could compound this acceleration, potentially reaching a point where its advancements outpace human understanding and control. The report highlights a tension between the practical application of advanced AI for faster development and the long-term governance challenges this speed presents.

The findings emerge despite Anthropic CEO Dario Amodei publicly urging the broader AI industry to slow its progress. Amodei's stated aim has been to ensure that safety measures can adequately keep pace with technological advancements. The report, therefore, points to a potential divergence between the company's internal development momentum, driven by effective AI-powered processes, and its leadership's external calls for a more cautious industry-wide approach.