By Islam El-Sakka
Anthropic's artificial intelligence models are advancing at double the pace recorded a year ago, according to an independent assessment by the non-profit Model Evaluation and Threat Research (METR). The findings, published Thursday, indicate the company has effectively achieved two years of benchmark progress in a single year.
Recursive Self-Improvement Drives Acceleration
The accelerating development stems from recursive self-improvement within the AI models themselves, rather than increased computing power for training. Researchers warn that if AI systems further accelerate their own development, it could lead to a sudden emergence of extreme and risky capabilities.
Amodei Calls for Paced Development
Anthropic CEO Dario Amodei has urged a slowdown in AI development. In a September essay titled "We Must Pace the Frontier," he proposed a three-point plan for safe development, noting that AI's growing ability to build the next generation of AI could outrun human understanding and control.
Industry Leaders Echo Concerns
OpenAI CEO Sam Altman agreed with Amodei's call to "pace the frontier" and endorsed the proposal for independent evaluators. Anthropic invited METR researchers to review internal data and evaluate model capabilities.
Human Oversight Still Critical
The METR report notes humans remain essential for overseeing the most critical parts of model training, though researchers observed "signs of life" suggesting AI could soon make autonomous contributions to important areas, with varying maturity across domains.
