因为会被更强的通用模型大白 ## Specialization loses out to systems with more compute and data The bitter lesson: our intuition that specialized human knowledge is better than scale is wrong. Richard Sutton: simple systems with more data and computing power always beat specialized human knowledge. This already happened in computer vision in the 2010s. It's happening now with AI writing tools—base models like ChatGPT made them obsolete. The same fate awaits ==specialist AI tools in finance, legal, and coding==. Why? Because general AI models are improving fast. They can handle tasks twice as long every seven months. **They'll eventually swallow any specialized tool built on top of them.** > If scaling laws hold and help models close the gap on domain expertise and specialized features, what looks like a durable edge today may just be temporary arbitrage. --- 但利用 AI 独特属性创造全新工作流的垂类 AI 有巨大机会 [[With new technology, invent totally new workflow and avoid skeumorphism]]