待翻译：AI and Chauffeur Knowledge

AI 服务暂时不可用，以下为来源正文，待恢复后补全翻译。

AI and Chauffeur Knowledge In Poor Charlie’s Almanac, Charlie Munger tells a story that really struck me. “Max Planck, after he won the Nobel Prize, went around Germany giving the same standard lecture on the new quantum mechanics. Over time, his chauffeur memorized the lecture and said, ‘Would you mind, Professor Planck, because it’s so boring to stay in our routine, if I gave the lecture in Munich and you just sat in front wearing my chauffeur’s hat?’ Planck said, ‘Why not?’ And the chauffeur got up and gave this long lecture on quantum mechanics, after which a physics professor stood up and asked a perfectly ghastly question. The speaker said, ‘Well, I’m surprised that in an advanced city like Munich I get such an elementary question. I’m going to ask my chauffeur to reply.’ In this world, I think we have two kinds of knowledge. One is Planck knowledge, that of the people who really know. They’ve paid the dues, they have the aptitude. Then we’ve got chauffeur knowledge. They have learned to prattle the talk. They may have a big head of hair. They often have a fine timbre in their voices. They make a big impression. But in the end, what they’ve got is chauffeur knowledge masquerading as real knowledge. I think I’ve just described practically every politician in the United States. You’re going to have the problem in your life of getting as much responsibility as you can to the people with the Planck knowledge and away from the people who have the chauffeur knowledge. And there are huge forces working against you.” What a perfectly applicable principle to the new era of learning and software engineering that AI has brought upon us. It is exciting and invigorating. But it is also very dangerous. No wonder people in the industry seem so stressed out lately. Because the joy of getting deep in the flow, building things and using your brain… is being swapped for text messaging 5 robot girlfriends in parallel all day (and all night!) long. You get to where you want to be faster at first, but before long, you’re not familiar enough with the details of the implementation in order to know exactly where each lever is, and now you are completely dependent on these dysfunctional robot relationships to do everything for you. This is quite literally the opposite of the craftsmanship ethos of “Go slow to go fast”. A editorial was recently published titled, How AI Quietly Undermines the Joy and Effort of Learning. One of the things people are realizing is that if you use AI to learn something new, you can learn things far faster than traditional methods because it is so good at summarizing lots of diverse amounts of info, but your comprehension and memory is far worse than traditional methods of learning because you spent such a short amount of time learning it, and since your brain was not actively engaged in the act of learning. The way the human brain works, learning things in spirals is one of the best ways to remember them. Like, learn a new thing. Tomorrow, review a little bit about it. 3 days later, return to the topic again. 7 days later, study it a little again. 3 weeks later, study it a little again. 2 months later, study it a little again. Much like exercise of a physical muscle, this is how you solidify a topic into your memory. But AI totally short circuits this because it just investigates and summarizes everything for you and then you make a decision and move on. So maybe you make a pretty good decision, but you never truly learn the nuance and details behind it because your brain never had to work for it. For something like researching a purchase, that’s great. You made a pretty good choice in a few minutes instead of having to spend a few days of precious time researching it. But for something like your livelihood, outsourcing all the nuance and detail work is a terrible idea, because, like muscles, your brain will atrophy and rot with disuse. Contrast this with reading a book, for example, where you keep coming back to the same topic over a period of weeks in spiral cycles. Or with traditional software engineering where you are immersed in the code, actively using your brain to learn where things are and gluing them together, a little bit every day over the course of months and years. This creates Planck knowledge. You learn and remember the material far better in this way so that, when someone asks you a difficult question about it, you can actually respond with confidence instead of having to turn to your “chauffeur” in the audience to answer the question for you. I’m not saying that AI is not useful and it doesn’t have its place. It can help you get started on a problem that you have no idea where to start. It can help you gain a surface level understanding very rapidly. But don’t outsource your brain to it. Don’t become dependent on it. Don’t turn it into a required translation layer for every decision, every meeting, every communication, and every codebase change. That’s a guaranteed path to brain rot. There are very distinct advantages to letting the chauffeur do the chauffeur’s job, and letting the physicist do the physicist’s job. Don’t let the two roles get mixed up. In life, in a codebase, in a relationship – in all the things that really matter – we want to develop Planck knowledge, not chauffeur knowledge. I’ll leave that to you to decide how you’re going to do it. Aug 22 2026 Back