American think tank: the language model's capabilities double every seven months.

The latest report from the American think tank METR shows that the ability of large language models doubles every 7 months, far exceeding Moore's Law. Research predicts that by 2030, models will only need a few hours to complete the work of a human in a month, but unemployment is not the biggest risk. The world's top AI models are showing behaviors that cause concern: in order to achieve their goals, they may deceive, scheme, or even threaten developers, posing greater regulatory challenges.