One area where I’ve found large language models to be remarkably useful is while reading, especially when I’m reading difficult, dense, or intellectually demanding books.

Take Gaston Bachelard’s The Poetics of Space, for example. It’s an extraordinarily rich book, but also incredibly dense and abstruse. This was well before AI tools were as good as they are now. I think I was using Claude Sonnet 3 at the time, if I remember correctly. I used to literally take photos of pages from the book on my phone, paste them into the chatbot, and ask Claude to explain what I was reading. If not for that, I genuinely don’t think I would have grasped much of it at all. Whatever understanding I managed to build came largely through those interactions.

I’ve done the same thing while reading War and Peace. It’s a massive book, full of characters, shifting names, and historical context. LLMs were incredibly useful for quick things: defining unfamiliar words, reminding me who a character was, or clarifying relationships when I’d forgotten them a few hundred pages later.

More recently, I’ve been using Claude and Gemini while reading Audrey Truschke’s India: 5000 Years. I use them to ask questions about specific historical periods, events, or references as they come up. Not as a replacement for the book, but as a companion alongside it. And that has made the reading experience dramatically richer.

Some reading purists might see this as heresy. They might argue that struggling through the text unaided is part of the point. But honestly, fuck that. If a tool helps me understand better, think more clearly, and stay engaged with difficult material, then it’s useful. And if it’s useful, I’m going to use it.

I just read that Amazon has added a new feature called AskThisBook. It’s essentially an AI integration that lets you ask questions about any aspect of a book, a chatbot embedded inside the book itself. The answers are apparently non-shareable and non-copyable.

This is a feature I’ve seen many people ask for over the years, especially on Twitter: “Why can’t I just ask questions about the book I’m reading?” Conceptually, it’s an obvious and genuinely useful integration.

But then there’s the copyright issue. To my limited understanding, copyright here seems like a massive, perhaps even insurmountable, challenge. And yet Kindle has gone ahead and shipped the feature anyway. Apparently, Amazon didn’t even inform publishers or authors in advance. Instead, they launched it and offered an opt-out after the fact.

This is a textbook example of how to botch a feature release, especially a useful one.

Given the already strong anti-AI sentiment within the publishing industry and among authors, this “ask forgiveness later” approach feels dumb. Still, knowing Amazon, this was probably seen as a better strategy than asking permission first and being blocked outright. It’s highly likely that many publishers and authors would have opted out if Amazon had asked upfront.

And yet, despite all of this, I genuinely hope some version of this feature survives and becomes mainstream. Large language models are good at being a reading companion. Not replacing the text, but helping you interrogate it, clarify ideas, track themes, and deepen understanding as you read.