Using A.I. for eBook summaries

[haertig]

For my use - refreshing my memory on books I've already read - I could live with some minor errors or discrepancies. Not too many. But some. Because my whole use is to jog MY memory, not replace those faint memories with something A.I concocted.

It sounds like A.I. summaries may not be so good for other uses - like if didn't want to read a book for a college class and decided to use an A.I. summary as the basis for homework you'd submit to your professor. ...but since the professor would probably be using A.I. to grade your homework, you might get away with it! What a world we live in - it's going to be a complete fantasy devoid of any reality by the time our generation croaks.

[ZodWallop]

Originally Posted by Readirect

I once asked AI to recommend 5 thrillers which critics praised but had low sales... Except 3 of the 5 didn't exist.

This may have happened. But it sounds an awful lot like this story from almost exactly a year ago: How an AI-generated summer reading list got published in major newspapers

[ZodWallop]

I don't really have an issue with an AI syopsis of a book. Seems like the sort of thing they are good at. But there's two issues I see:

1. In 2019, I bought the book Cold Storage by David Koepp. I bought it because I've really enjoyed the movies he's written and directed. But I haven't yet read the book, because the synopsis just didn't grab me. Now the book has been made into a movie and the revised synopsis sounds much more compelling. Writing a synopsis is easy. Writing a good one is a whole different thing.
   
2. There's been a huge increase in AI generated content on YouTube. When I've purposely tried watching a long form AI generated video (say, something over eight minutes or so) I just can't make myself sit through it. AI isn't good at making something compelling and starts to repeat the same points over and over.

As a publisher, I wouldn't want AI to generate one of the things most likely to help a customer make a decision on whether to purchase a book or not.

As an end user, I wouldn't want an AI summarize the book as it may well hallucinate and if the book is decently popular, it is just as easy to go to Amazon, Goodreads, etc. and find a better written synopsis.

[SteveEisenberg]

Originally Posted by Readirect

I once asked AI to recommend 5 thrillers which critics praised but had low sales.

AI is imperfect but hallucination seems to me a lot less than last year.

I just tried this query at claude.ai, and they correctly report that "publishers rarely publicize poor sales figures."

[DNSB]

Originally Posted by ZodWallop

This may have happened. But it sounds an awful lot like this story from almost exactly a year ago: How an AI-generated summer reading list got published in major newspapers

I know several people who tried that experiment when that article was first published and regardless of criteria or which of 5 AI providers used, on a request for 10 recommended books, the average was 1.5 hallucinations. Perhaps the AI was accessing the contents of The Library of Lost Books from Holly Lisle's book Hell on High?

In March, 2026, one lawyer representing an indigenous fisheries agency filed a request for more time to file for a judicial review. Sadly, the 4 court precedents cited in the request turned out to be AI hallucinations. Despite the number of reports about issues with AI being used to supply precedents, there was no effort to check if the precedents actually existed.

[GoodRiddance]

AI can do fine summaries.

The main problem is that what catches people, is emotion. AI is very bad at catching emotion when it isn't explicit.

You have these books that describe situations and you can 'feel' what happens but it isn't actually being said specifically.

AI will definitely miss that.

And to be honest, exactly that is what makes good books. Not the story, but the feeling you get when reading the story. Good writers can do that. They can even make average stories great to 'experience' by reading.

[pickyaxe]

this thread contains two cookie-cutter takes that I've grown very tired of reading in my circles:

1. LLMs are fancy pattern matchers and therefore aren't ever useful in real-world tasks. also they're killing the planet.
2. frontier models do not have this problem, you must have used a model from X months ago.

[DNSB]

Originally Posted by pickyaxe

this thread contains two cookie-cutter takes that I've grown very tired of reading in my circles:

1. LLMs are fancy pattern matchers and therefore aren't ever useful in real-world tasks. also they're killing the planet.
2. frontier models do not have this problem, you must have used a model from X months ago.

• LLMs are useful for some purposes but require supervision for most tasks. Sadly, that supervision requires a human with enough expertise to be to decide if the output is valid. As for killing the planet, look at the power requirements which are currently ~5% of the total North America generation capacity. You did notice that Meta and Microsoft are looking at firing up new nuclear generation stations? Is the amount of power consumed by AI data centers going to drop? All the sources I've seen say that is very unlikely.
• Sadly, even with the latest models I've tested, the results are jagged. There are improvements but still not at the point where I, personally, would want to use them as a sole source.

[JSWolf]

I've seen some pretty dire AI videos on YT.

[JSWolf]

I asked Copilot (website)...

I want a list of 10 science fiction books available as eBooks for summer reading.

1. Project Hail Mary — Andy Weir
2. Dune — Frank Herbert
3. The Hitchhiker’s Guide to the Galaxy — Douglas Adams
4. The Martian — Andy Weir
5. Neuromancer — William Gibson
6. The Left Hand of Darkness — Ursula K. Le Gui
7. The Three‑Body Problem — Cixin Liu
8. Ancillary Justice — Ann Leckie
9. Station Eleven — Emily St. John Mandel
10. The Long Way to a Small, Angry Planet — Becky Chambers

All are real books.

[Tawn]

Originally Posted by haertig

I am not an A.I. user and I don't know much about its capabilities. I was wondering about A.I. and eBooks. Has it advanced to the point where you can give it an EPUB and it will come up with a plot summary of the book (I'm talking about fiction books)? That would be useful to me when going to read a new book in a series where I haven't read earlier books in the series in a long time. In other words, summarize books I've already read so I can catch myself up on important details before continuing the rest of the series...

I'm guessing you'd probably get enough information to be at least somewhat useful. Give it specific details (e.g. I've read this previously and need a refresher before reading the next book in the series). I've done this with a book club book I'd read previously and it did a great job.

OTOH I've gotten some nonsensical responses on other topics like 'add water to x to make it thicker', impressive looking code (that doesn't work), and BS like 'just do x' (that's not possible)...'you're absolutely correct; you need to do y' (that doesn't work)...'you're absolutely correct' (oh, never mind...).

Still I'd think if you'd previously read the book, you'd likely notice any errors should there be any. AI certainly isn't 100% trustworthy, but neither is internet content in general.