21 September 2025 · 2 min
When I was an undergraduate student, I discovered symbolic algebra. It was great! Instead of solving for a variable by hand, I could just put all the equations in the machine and get the result.
I soon found that symbolic algebra did not turn me into a genius mathematician. It could do the boring work, but I often found myself “getting stuck.” I would start from a problem, throw it at the machine and get back a mess that did not get me closer to the solution.
Over time, I realized that these tools had an undesirable effect on me. You see, human beings are lazy by nature. If you can avoid having to think hard about an issue, you will.
And I fear that much of the same is happening with large language models. Why think it through? Why read the documentation? Let us just have the machine try and try again until it succeeds.
It works. Symbolic algebra can solve lots of interesting mathematical problems. Large language models can solve even more problems.
But if you just sit there and mindlessly babysit a large language model, where are your new skills going to come from? Where are your deep insights going to come from?
I am not being a Luddite. I encourage you to embrace new tools when you can. But you should not dispense with doing the hard work.
Daniel Lemire, "Hard work is a virtue," in Daniel Lemire's blog, September 21, 2025, https://lemire.me/blog/2025/09/21/hard-work-is-a-virtue/.
[BibTeX]
Deskilling appears to be a very real problem with LLMs.
Because you have the AI do the simple stuff, you lose the skills to do the simple stuff yourself, and hence also to do more complicated things
The best parallel to that problem I’m aware of is physical labor, all kinds of heavy machinery and cars and so on made us capable to do a lot more work than before, in the same time making us physically weak. But we since realized that fitness is great for us and we (well, some of us) voluntarily spend time exercising. And some even do sports for no other reason but to compete in useless exercise.
The mental work will have the same fate: some will degrade, some will exercise for the joy of it and others will exercise for the competition.
Where did the lower half of the man in the hammock’s body go!? That’s going to give me nightmares.
Slow work has inherent value. A part of the problem is that we get rewarded too much for the product (and getting there as quickly as possible) and not enough for the integrity of the process.