#AnthropicCEO呼吁放缓AI发展 People say it’s “slowing down,” not stopping AI—it’s about setting the pace for cutting-edge capabilities: models can already write code, run agents, and help build the next generation of models, but alignment/evaluation hasn’t caught up;
Look at it in three layers:
Technical: recursive self-improvement + multi-agent collaboration shift the risk from “saying the wrong thing” to “autonomous action / attacking systems.”
Business: Anthropic sells “the safest frontier models,” turning safety into a brand and a moat; third-party on-site evaluations benefit big tech, while startups/open source are harder hit.
Geopolitical: if the US unilaterally stops but China doesn’t, it’s a loss; so he calls for slowdown while also supporting chip regulation and urging coordination among democratic countries first.
In short, it’s not just apocalyptic marketing, and it’s not purely altruism—it's the convergence of rising risk + incoming regulation + big firms racing to define the standard.
Look at it in three layers:
Technical: recursive self-improvement + multi-agent collaboration shift the risk from “saying the wrong thing” to “autonomous action / attacking systems.”
Business: Anthropic sells “the safest frontier models,” turning safety into a brand and a moat; third-party on-site evaluations benefit big tech, while startups/open source are harder hit.
Geopolitical: if the US unilaterally stops but China doesn’t, it’s a loss; so he calls for slowdown while also supporting chip regulation and urging coordination among democratic countries first.
In short, it’s not just apocalyptic marketing, and it’s not purely altruism—it's the convergence of rising risk + incoming regulation + big firms racing to define the standard.