[al_borland]

Books exist for people who want in depth information with the full context, in an organized manner.

Short forms have always been available, it be blog posts, Wikipedia articles, cliff notes, or other such things. Books survive, because source material is needed to generate all of those other things, and those short form versions don’t cut it for everyone. I don’t see LLMs as any different.

A book can tell you something you didn’t know. With an LLM you need to know enough to ask.

[kadushka]

With an LLM you need to know enough to ask

Not really. You can say "I'd like to learn about X", and get a high level overview about the topic you're interested in. Then just ask more specific questions, just like you would do when interacting with a human teacher. Try it with Claude or 4o.

[skydhash]

That seems very inefficient. And dangerous as it's well known that LLMs will hallucinate. When I want to Learn about X, I just take an introductory book, look at the table of contents and the description to know what it covers, then just read it. I then have several paths I can take to further my knowledge.

> Then just ask more specific questions, just like you would do when interacting with a human teacher

And no one interacts like that with a teacher, at least from what I know. They have a syllabus for a reason, and while they may answer an unrelated question, they expect you to master the fundamental that leads to it. If you don't they will point you to it, because the answer will be difficult to understand. It's like wanting to know how nuclear works without learning atom composition first.

[benrutter]

Yeah I completely agree with this! I really like books for learning because they do exactly this.

Take the Rust book, you have a neat and organised collection of the majority of Rust features you're likely to use.

An LLM might be handy in answering questions like "Why is the borrow checker failing this code?" but that's a really different proposition to getting a detailed and complete summary of Rust's key features. It could maybe output something along these lines, but I think the output would be considerably less usable and reliable than a book.

[skydhash]

The moment you gain some expertise in a subject, LLMs fail horribly. Because you will have a mind model of the thing you're working with, LLM won't be able to solve what you can't, as it will often require a deep dive into the internals. And that's when you want a complete reference/manual nearby. As for boilerplate, most experts have project templates or can extract one from an old project.