告诉HN：一致性无法扩展
自从去年11月辞去软件工程师的工作，因为我的老板推动不负责任地使用人工智能，我花了大量时间试图弄清楚人工智能在未来经济中的实际位置。<p>多亏了一些出色的Substack作者[0,1]，以及代理工程正在重新发现我们关于人类委托的所有知识，我终于有了突破：一致性是无法扩展的。<p>大量的精力已经被并将继续被投入到试图克服这一简单但可能是普遍适用的法则中。<p>这一法则还保证了总会有工作在扩展那些一致的事物，以及将那些已经扩展的事物进行一致化。<p>0: https://thestrategiclinguist.substack.com/p/the-rule-that-never-was-what-7-38
1: https://abiawomosu.substack.com/p/so-they-built-an-ai-that-undresses
查看原文
Since quitting my software engineering job in November because my boss was pushing reckless use of AI, I've spent a ton of time trying to figure out what place AI actually has in the future economy.<p>Thanks in large part to some amazing Substack authors[0,1] as well as the fact that agentic engineering is in the process of rediscovering everything that we have learned about human delegation, I finally had a breakthrough: coherence doesn't scale.<p>A ton of energy has been & will be expended trying to overcome this simple, yet possibly universal law.<p>This law also guarantees that there will always be work scaling things that are coherent & cohering things that have scaled.<p>0: https://thestrategiclinguist.substack.com/p/the-rule-that-never-was-what-7-38
1: https://abiawomosu.substack.com/p/so-they-built-an-ai-that-undresses