Definitely isn't the latter. Numerical Recipes and Hacker's Delight have tons of gems that you won't get from an LLM, or that an LLM will even understand despite appearing all their training sets.

Even before LLMs became big I started hording solid technical books, as there was so much misinformation on Google/SO that any non-trivial technical question could not be answer without a high probability that the answer was fundamentally wrong.

LLMs are super helpful for learning, but without the foundation of a true textbook at your side they will very easily go off the rails into a world of imagination.

Any recommendations for a few keepers a programmer should have?

I think it’s largely because of eBooks, and the proliferation of lower-effort eBooks on the market. The latter is increasing because people use LLMs to write them. To actually make money on tech books, you kind of have to be a juggernaut or have some runaway hit titles. Once isn’t enough, either, because someone’s always putting out something newer on the same topic or people are moving to a different stack altogether.

even before LLMs you could/can get a lot of info on the web. Most ideas presented in books exist on the internet somewhere, quite a few pass through HN. In the 90s and early 2ks I used to hoard books, but now, not so much, I get a few through my library these days but a lot of times I just find the book is padded out to be a book and the meaty bit of the book is usually tiny.

I also find that docs have gotten a lot better over time as well. I feel that Y2K era of numerous large tomes was largely to fill in the gap left after software publishers stopped shipping even their full price boxed software with meaty manuals, assuming I guess that help menus and GUI context (or maybe even Clippy) would get you there. Even some of the enterprisey software- Oracle comes to mind- would ship with like 6-10 volumes of reference information but very little in the way of a getting started guide that showed you how to get basic stuff up and running or what best practices were.

The web got a little better, and what drove brainshare and usage was a good experience getting started with reasonable defaults and good docs to get you started. API design is also much better these days- I was trying to find some examples of how unintuitive say MFC (Microsoft Foundation Classes- big in the win9x days) apis were, but a lot of those docs seem to have disappeared- here is a stackoverflow/experts-exchange links that show how non-intuitive it was: https://stackoverflow.com/questions/3255207/window-handle-in... https://www.experts-exchange.com/questions/10018203/How-to-g...

Motif on the Unix side of things was a bit better, but not much. You really needed a good book to walk you through things in a more understandable way.

Another option might be that people are increasingly using LLMs to write the books.

This is it. I know of too many companies (in nonfiction, specifically) trying to offload as much as possible to LLMs, damn the quality. Zero chance I buy any nonfiction written after 2021 or so unless it comes with strong recommendations from sources I trust. No more on-a-whim purchases because something looks interesting and is on sale or whatever. That’s over.

But even before LLMs, how many of those nonfiction books were any better than an assembly of random blog snippets?

Damn few of them.

I'm thinking the "industry" will contract a lot, but there will always be a niche for deep "real books" on a subject. It will just be much smaller.

loaded questions.

Maybe a bit depressing, but I'm not sure loaded. I was just talking to someone (in person!) a few days ago, who purported that online courses were basically dead, because people can learn from LLMs instead.

And then, it seems to be a real issue amongst some people to ask, "why should I learn X, when LLMs already know it?" Not unlike, "why should I learn to divide, when we have calculators?" but on a grander scale.

Statements like these always brings me to memory the opening line of Hamming's Numerical Methods book: The purpose of computing is insight, not numbers. It is very easy to get carried away and forget that - in particular today when processing power grows exponentially. Even more when we know there are a myriad of problems that are uncomputable, literally, and human common sense and intuition (insight) are as relevant now as ever.

My hypothesis is people will get burned out on this unguided learning via LLMs and still want some sort of curated/guided learning experience through material to understand some subject.

There is the problem of "I don't know what I don't know" that a course can solve for you. An LLM can sort of do that, but you have to take its word for it, and it does it pretty much strictly worse at the moment (but is much more flexible).

> My hypothesis is people will get burned out on this unguided learning via LLMs [...]

I'm less optimistic. Already 20+ years ago many people complained if you pointed them to books which answered their questions in depth. The standard reply was "just tell me how to solve this particular problem" instead.

<cough> StackOverflow... they made a business out of that.