Show HN: Book rewrite (reduction of 40-60%) service with LLMs
Hi, i've built this for myself, to be able to read more, in less time. It's a 40-60% reduction rewrite, not a summary. My aim was quality. Here's an example with Walden with a 53% reduction: https://file.io/tA5jmkFgCz9d
9 comments
Interesting, why would you summarize the book, if the whole point of reading it, is reading it?
A man once told me, if the book is boring to you, just leave it, no need to read it full. Not every book is interesting, turns out.
The other case, if I find boring parts within the book, I just skip them myself.
In the end, give the state of LLM, I would be afraid that I'll miss some interesting parts or get things that are not true.
What I think LLM would be useful is excatly the thing you said its not - summary. I would love to get a summary bullet points every chapter or every few pages. That would be really useful.
In the end, give the state of LLM, I would be afraid that I'll miss some interesting parts or get things that are not true.
What I think LLM would be useful is excatly the thing you said its not - summary. I would love to get a summary bullet points every chapter or every few pages. That would be really useful.
Summaries are easier to generate, but making it shorter, still having the author's voice, is more appropriate for my goals. I've tried a lot of summary iterations and it turns out i don't want to read ABOUT the text.
It's still reading it, just by the time you get to 50% of the book, you could have finished it, with more details than a summary. A summary can be as short as 5-20%.
My description does not capture what the output is really, check the rewrite and see if you miss anything critical.
It's still reading it, just by the time you get to 50% of the book, you could have finished it, with more details than a summary. A summary can be as short as 5-20%.
My description does not capture what the output is really, check the rewrite and see if you miss anything critical.
Updated link: https://filebin.net/29yn8bqtq6cx34ii
Hey, what was your process for making this? Any insight on what are the things it sacrifices on?
Will you share the prompt or model or setup you used?