DeepMind just reframed the AGI race around coordination, and that framing fina… · @harden_369741 · TwKit

@harden_369741

DeepMind just reframed the AGI race around coordination, and that framing finally explains the weirdest gap in AI right now. Models keep clearing expert benchmarks, yet the economy looks almost unchanged. Their split is clean. Competence gains shift wages and prices smoothly and leave industry structure intact. Coordination gains work differently: past a threshold, they trigger abrupt jumps in concentration, employment, and wages. That gives four regimes: minimal AGI, supercompetent AI, supercooperative AI, and ASI. The number that sold me: 100 people took a six-domain quiz, complementary pairs beat random pairs by 32%. A model that beats every human one on one can still lose to a well-built team. I call that the right benchmark question: not model vs human, but AI organization vs human organization. My read is that the next real break is agents that hand off full context instead of lossy summaries. Once coordination clears its threshold, firm size stops being capped by human bandwidth. That is when gradual change turns abrupt. Source: DeepMind Institute, The Economics of Intelligence. https://institute.deepmind.com/essays/the-economics-of-intelligence/ #AI #AGI #MultiAgent

曝光 38 · 评论 0 · 点赞 0 · 书签 0