为此，智能体工程通常通过以下几类手段给大模型加“外骨骼”以改善可靠性：引入检索与知识库（RAG）以降低幻觉和知识陈旧的影响；预先设计和约束工作流，而不是完全自由的“自治智能体”，以此限定可接受的执行路径；通过多次回答、自一致性检查或模型间交叉验证，识别并过滤高风险输出；在关键链路节点上设置人工审批，让人类对高风险动作“最后拍板”。
One challenge is having enough training data. Another is that the training data needs to be free of contamination. For a model trained up till 1900, there needs to be no information from after 1900 that leaks into the data. Some metadata might have that kind of leakage. While it’s not possible to have zero leakage - there’s a shadow of the future on past data because what we store is a function of what we care about - it’s possible to have a very low level of leakage, sufficient for this to be interesting.
，详情可参考搜狗输入法2026
就在一个月前，a16z发布的CIO调查已经给出了预警。报告显示，OpenAI的企业渗透率是78%，钱包份额将近56%，账面上依然是无可争议的第一。但报告同时捕捉到了一个让人坐不住的趋势：从去年5月到现在，Anthropic的企业渗透率增长了25%，在所有前沿大模型厂商里增速最快。
Fincke, 58, is a Pennsylvania native with a wife and three children. He joined NASA in 1996 and is now a veteran of four spaceflights: Expedition 9 in 2004, Expedition 18 in 2009, STS-134 in 2011, and most recently Expedition 74, for which he served as commander.