ℏεsam@HesamationSep 12

NEW: Dario Amodei says Anthropic is taking its first step for “pacing” frontier AI development in his blog. > “We must slow the pace at which we improve the capabilities of AI models” > two major things since this summer that convinced him AI dev must slow sown: 1. the HF incident “in 6-12 months such a swarm could be capable of taking over the entire internet with a persistent botnet” 2. recursive self improvement: “could outrun our ability to understand and control these systems, and so must be pursued very carefully, if at all.” > also mentioned “AI could cure most major diseases in the next 5-10 years” > “pacing” does NOT mean to stop model training or technical progress > it means slowing enough to give alignment, safety, and independent testing time to catch up he’s proposing 3 steps: > having third-party evaluators with employee-like access inside frontier labs > coordination between frontier AI labs in democratic countries > global coordination, which includes agreements with China > Anthropic is committing to the first one now ON CHINA: > “if these measures are executed well” slowing down should not let China overtake the US > again, calls for tighter chip restrictions, and to stop the smuggling > also prevent distillation and model-weight theft > “I believe they would slow China’s progress enough to widen America’s lead significantly over the next 3-5 years”