请问HN：大型语言模型（LLMs）是否正在逐渐使公司变得功能失调？
大型语言模型（LLMs）非常出色。它们能够快速提供许多问题的合理解决方案。这使得公司积极推动将日益增长的认知需求转移到LLMs上。
投入精力去验证一个看似合理的、由LLM审查和简单测试的临时解决方案，并不太有意义，尤其是在疲惫的人类只需快速浏览一下的情况下。
这导致了各个方面的复杂性不断增加：决策、沟通、代码、文档等等。
与认知债务相当，随着思考和决策的外包，没人获得解决特定细分用例的独特经验。这就像一个知识渊博的人在知识传递之前离开公司。你所剩下的只是别人假设和决策的遗留，且你永远无法完全了解。
我已经看到了这种情况。在支持方面，LLMs从未解决过我的任何支持请求。支持团队本应对产品了如指掌，却推荐访问一些虚构的页面和按钮，而这些在产品中并不存在。在代码方面，代码库的复杂性不断膨胀，变得不合理。在产品方面，界面决策被外包给LLMs，使得公司在面对独特的细分问题时得出相同的解决方案。
你是否也观察到了类似的现象？
查看原文
LLMs are great. They can quickly one-shot plausible solutions for many problems. This makes companies push hard to offload rising cognitive demands to LLMs.<p>It doesn't make much sense to invest effort into verifying a plausible-looking throwaway solution that is reviewed and smoke-tested by an LLM, with a quick glance from a tired human.<p>This results in complexity creep everywhere: decisions, communication, code, documentation, etc.<p>On par with cognitive debt, as thinking and decision refinement were offloaded, nobody gained unique experience solving a niche use case. This is equal to having a knowledgeable person leave the company before the knowledge transfer occurred. All you're left with is a legacy of someone else's assumptions and decisions you'll never be fully aware of.<p>I already see it happening. In support, LLMs never solved any of my support requests. Support, which should know the product inside out, recommends visiting pages and clicking buttons that are hallucinated and don't exist in the product. In code, codebases are ballooning with unreasonable complexity. In product, interface decisions are outsourced to LLMs, making companies arrive at the same solution for their unique niche problems.<p>Are you observing something similar?