Hyper-Reading
How I read and learn from books with AI
Some books are a pleasure to read. All I want is to sink into the couch and spend some quality time with them, reading and re-reading, savouring every word. Most books, especially contemporary non-fiction books, are not pleasurable at all. Many of them, even while being useful as vessels for information, are awful. I read them to expose my mind to new information and try to retain some of it in the hope that it will help me know stuff and think interesting things.
Over the years I have tried different approaches to working through such books, but AI has completely transformed how I read and learn. This is my current (Autumn 2024) recipe - Hyper-Reading:
Acquire the book in unrestricted textual formats. Ideally both EPUB and PDF.
Explaining how to do this is beyond the scope of this document. Suffice to say that there are no strictly technical limitations to achieving this.
Upload the PDF to NotebookLM.
NotebookLM does not read EPUB files. If I don’t have a PDF I might use a tool like Pandoc to convert the EPUB into a format that NotebookLM accepts, like an HTML file or a Google Doc.
Produce and review the standard notes suggested by NotebookLM (brief, study guide, FAQ, …) and if I’m not too lazy think of a few other notes I can get it to produce that will help me grok the book.
Read the notes carefully. At this point I want to have a good understanding of the structure of the book and the topics covered. I am now primed for learning what the book is ready to teach me.
Produce a NotebookLM “deep dive” podcast.
Now that I’m “primed” with the structure and topics covered, I can and do customise the podcast to focus on things that are interesting to me.
Some things I almost always ask for in the customisation: quote from the book (at least 3 quotes for each topic covered), focus on the generalisable topics covered rather than on anecdotes (”business books” are often full of boring anecdotes and the NotebookLM podcast loves to dwell on them), highlight important facts and concepts I should remember, highlight actionable insights, etc….
Listen to the deep-dive podcast.
Alternatively, in some cases when I think it will be worth my time, I get the audiobook and listen to it while I go for a walk, lie in bed failing to fall asleep, or do chores. Most books are not worth it, but some are, especially when audiobooks are produced well (some contemporary audiobook productions are really fun).
Reflect on what I know so far. Consider what I want to understand better.
Skim the book. I might read whole chapters or just browse through a couple of pages. This can take anywhere from a few minutes to a few hours, depending on how engaging the book is. Most non-fiction books are not really worth the time and I trust my instincts.
(optional) chat with NotebookLM to go into more detail on some topics. Or maybe produce some additional notes.
Upload the EPUB to MkFlashcards and use it to produce flashcards for spaced-repetition learning.
I over-generate. MkFlashcards is very simple and still doesn’t have good mechanisms for focusing on what’s important and ignoring what isn’t, so I generate 3 times as many flashcards as I think I probably want to review. I will later delete many of them.
As the developer of MkFlashcards, this when I usually figure out some things that need to be improved, so I log issues in Github.
Import the flashcards into Mochi.
Review each card carefully. If I have a few hundred cards, which is typical, I might do this in installments over a few days.
If I don’t think the card is something I want to remember, I delete it immediately. I typically delete 2/3 of the cards I imported.
If I don’t understand the card but think it’s something I should understand, I go back to NotebookLM or to the book and learn more.
If the card is interesting but written or structured in a way that isn’t optimal for me, I edit it. This is quite common. Sometimes I’ll just highlight something, other times it’s a complete rewrite.
That’s it. I will now be reviewing these cards repeatedly as they get resurfaced by the spaced-repetition algo. Sometimes during reviews, I will feel the need to go back to the book or to NotebookLM. Doesn’t happen too often, but it’s a good instinct to follow, because it usually results in deeper understanding and better retention.
I have now read about a couple dozen books using this method, and it’s both a more engaging experience and results in more effective learning than anything I tried previously.
this is such a good workflow, thanks for sharing!!!!