人工智能巨头试图理解大型语言模型（LLM）与“代理工具”结合后的输出=理念
我认为OpenAI、Claude等对智能体的定义并没有被理解在同一个层面上。微软在萨提亚·纳德拉接管时曾经理解过这一点，但随后又放弃了，可能这也是他们收购OpenAI的原因。谷歌和苹果在他们的AI助手上摸索不前，这些助手仅能在手机上运行以安排日程。Meta可能稍微活跃一些，其开源项目显示了这一点。但没有一家大公司在AI问题上建立起自己的基础。
AI曾经被大型科技公司视为玩物或玩具，仅在内部使用或偶尔对外发布，而现在它的性质发生了变化。就像苹果和微软曾经在消费级计算机上争夺处理器性能、定价和设计一样，市场也定义了类别。如今，这些类别随着AI模型的非确定性答案或应用而发生变化。现实中有真正的客户和真实的需求，企业的产品没有保证的利润，而这是有史以来最灵活的技术。
在物理学中，一个元素的灵活性和抗压性使其适用于几乎任何机制或结构。AI就是这样的元素，而发现它的科学行业却无法真正弄清楚该如何调整它以适应哪些结构或机制。关键在于，大型科技公司自我安慰于其创建AI的项目，但它们并不是AI公司。它们是最被炒作的消费者，这正是现在向富人出售的最佳时机。
查看原文
I don't think the definitions of agents from OpenAI, Claude, and so on are being understood in the same spectrum. Microsoft understood it at some point when Satya Nadella took over, but then abandoned it, probably the reason they bought OpenAI. Google and Apple fumbled with their AI assistants that only ran on phones to schedule appointments. Meta is probably a bit more alive, as shown by its open-source projects. But none of the big players are building their foundations on the AI problem.<p>AI, once treated as candy or a toy by big tech companies and used internally or occasionally released to the public, is now something else. Just like Apple and Microsoft once fought over processor power, pricing, and design in consumer-level computers, the market had defined categories. Now, those categories shift with AI models with non-deterministic answers or application. There are real clients with real needs, companies with no guaranteed profits from their products, and the most flexible technology of all time.<p>In physics, the capabilities of an element being both flexible and resistant make it applicable to almost any mechanism or structure. AI is that element, and the scientific industries that discovered it can’t really figure out what structures or mechanisms to adapt it. The point is, big tech companies convinces itself of its project of creating AI, but they are not AI companies. They are the most hyped consumers of all, and that’s exactly why this is the right time to sell to the rich.