I’ve spent time to read the AI summary (which I normally never do) and the original (of course), and I can tell you the AI summary missed the most important point of the writing. I’d say what they are because it would defy the very purpose of what I’m trying to do.
But I can tell you that it’s enough to refute the AI fallacy many believe.
It's a shame grandparent's link gives a 404 now, I'd love to compare the two.
Surely this could be corrected by using the correct prompt? If you ask an AI to summarise a book, it's going to cut out 99% of the content. If you wanted to read some of that 99% of the content, you need to be clear about what you want. Otherwise you're leaving the AI to decide. That summary might be good enough for others, but not you. Tell it the focus you would like it to use. E.g. "include anything about the philosophy behind the decision." Whatever your point of interest.
To be clear, a summary will never replace reading 300 pages. But it might be 95% good enough for most people and in most cases, and that's often good enough.
Most of the time I’ve gotten major value or insight out of a book it’s because it’s said something truly unique or insightful that I genuinely never would have thought of if I hadn’t read it. Your prompt is useless because I might think I care about area A but I’d have my mind blown by area B which your prompt will drop unless it somehow has a perfect theory of my mind.
Sometimes you just have to read to damn book.
Agreed. There’s also something about “time under consideration.” The relatively slow pace of reading the book opens the door to personal insight that a summary hides. The words themselves might be important, but I’ve found the additional minutes of a book is critical for me to fully apply the concepts to my life.
If I know what the book contains to write a perfect prompt, why would I need a AI-summary?
I don’t say AI cannot summarize or there is no value in book summary, but I’d say that by stripping off most of a book, you almost certainly lose something valuable about the book, even if it is a prescriptive nonfiction. This most valuable part (besides the advices) is very often the part that actually convinced you and help the arguments stick. And without that, information will just stay as information.
This makes LLM summaries totally useless, then, surely? Without reading the thing, you cannot know which important bits to tell the LLM to retain; if it can't manage that itself, it's useless.
If a standard prompt gives you the wrong or spurious answer, the model is not a good model. In 2026, we should not be having to engineer a summarize prompt.