人工智能的风险并不是让我们变得懒惰，而是让“懒惰”看起来像是高效的工作。
我一直在思考大型语言模型（LLM）如何改变我们作为工程师的学习习惯，并意识到一个令人担忧的事实。
人工智能现在能够快速帮助我们搜索和研究信息，将论文的核心内容提炼成简明的摘要。这让我们能够迅速掌握一个术语，并有话可说。
但真正的学习需要深入阅读、思考和实践。一个精炼的摘要远远不够。自从有了人工智能以来，你有多久没有真正研究一篇论文或深入阅读并实施一项技术了？你的思维能力和品味是提升了还是下降了？一旦这种能力减弱，你准备好让人工智能完全取代你吗？品味不是通过阅读摘要建立的，而是通过无数次错误的决策和卓越的实践锻造出来的。
老实说，大多数人在有人工智能之前也从未认真读完很多论文。人工智能并没有剥夺任何东西——它只是让浅尝辄止的学习变得更加高效和更具欺骗性。真正的风险不是人工智能让人懒惰，而是让“懒惰”看起来像是“高效”。花十分钟阅读一个摘要，发到社交媒体上，感觉自己跟上了前沿——但实际上什么都没有留下。
我绝对不是反对人工智能。我提倡的是将人工智能用于深入工作，而不是把它当作假装学习的TikTok。从“为我总结”到“和我辩论”，从“为我做”到“帮助我推理”——这才是关键所在。
查看原文
I've been reflecting on how LLMs are changing our learning habits as engineers, and realized something worrying.<p>AI can now quickly help search and research information, distilling the core of a paper into a concise summary. It lets you pick up a term fast and have something to talk about.<p>But real learning requires deep reading, thinking, and practice. A polished summary is far from enough. Since having AI, how long has it been since you truly studied a paper or deeply read through and implemented a technology? Has your ability to think and your taste improved or declined? Once that ability is weakened, are you ready to let AI replace you entirely? Taste is never built by reading abstracts — it is forged through countless bad decisions and excellent practice.<p>To be honest, most people never seriously finished reading many papers before AI either. AI hasn't taken anything away — it has just made shallow learning more efficient and more deceptive. The real risk isn't that AI makes people lazy, but that AI makes "lazy" look like "productive." Spend ten minutes reading a summary, post it on social media, feel like you're keeping up with the frontier — but nothing actually sticks.<p>I am absolutely not against AI. What I advocate is using AI for deep work, not treating it as your TikTok of pretend learning. From "summarize it for me" to "debate it with me," from "do it for me" to "help me reason through it" — that is what matters.