omarsar0· @omarsar0 · X·· 3 天前AI 评分
智能体规模化下 RAG 研究需重审
顺便说一句，RAG 从未真正消亡。随着我们将智能体扩展到需要持久运行和不同检索格式的长时程任务，我们将需要重新审视这一切。只有经过规模化测试，你才真正知道某样东西是否经得起时间考验。早期关于 RAG 的预测是在智能体以小规模运行时做出的。我们开始看到，个人化且更自动化的多智能体系统为规模化智能体带来了更有希望的方向。我预计这里会有更多研究。
BTW, RAG was never really dead. As we scale agents into long-horizon tasks that require persistence and different retrieval formats, we'll need to revisit all of this. You only really know now if something stands the test of time when you test it at scale. Early RAG predictions were made when agents operated at small scale. We are starting to see a more promising direction for agents at scale with personal and more automated multi-agent systems. I expect to see a lot more research here.
来源：omarsar0 · x.com