我的观点原本是：只要你真的懂自己在干什么，验证 AI 生成的输出往往比自己写代码更慢、也更容易出错，因为 AI 的输出实在太啰嗦了。而如果你本来就不太懂，那用 AI 反而更不靠谱——你基本没法真正验证或事实核查它的结果，最多只能看看代码能不能跑（说白了就是在“凭感觉写代码 / vibe coding”）。
但现在我才意识到，AI 是可以定制成输出非常不啰嗦的，这样验证起来就容易多了。我完全不知道这点，真的有点离谱。毕竟我见过那么多人抱怨 AI 输出太花哨、太冗长了 ，你会以为大家早就在讨论这个了。
如果一开始需要验证的内容就更少，那我确实能理解为什么 AI 会提升生产力了。
My argument was that validating output from AI takes longer and is more error-prone than writing code yourself as long as you know what you're doing because output from AI is super verbose. And if you don't know what you're doing, using AI is even less advisable because you can't really validate or fact check it well enough other than just seeing if the code actually runs (you're basically vibe coding at that point). Now, I'm realizing that AI can be customized to have much less verbose output, which makes validation a lot easier. I have no idea why I didn't know about this. You'd think everyone would be talking about this given how many people I've seen complain about AI output being too flowery and verbose . If I can validate output faster thanks to there being less output to verify in the first place, then I can see how AI does improve productivity.
