Anthropic’s AI development has doubled in speed despite slowdown calls, research group says

Anthropic, a nonprofit AI research company, is developing its artificial intelligence systems at twice the speed it was a year ago, according to a report from the Berkeley, California-based Model Evaluation and Threat Research (METR). The findings come as Anthropic's CEO, Dario Amodei, has called for the industry to slow down the pace of AI model improvements to ensure safety measures can keep up.

Despite this, METR's research indicates that much of the acceleration in Anthropic's development is due to the AI models themselves, particularly Anthropic's own Claude chatbot, which is increasingly assisting in the development of newer models. The nonprofit, funded through donations, has gained unprecedented access to Anthropic's internal data and conducted interviews with the company's staff.

While Anthropic's redactions limited the depth of METR's analysis, the group believes the company's rate of improvement has doubled over the past year. The report also notes that Anthropic's progress is primarily in "well-specified tasks" that can be graded automatically, rather than more complex tasks. However, the researchers acknowledge that it's unclear whether this acceleration will continue, as there are potential bottlenecks such as restrictions on computing power or data that could limit further progress.