Service of Feeding Physical Books to AI

Continuing from nztm 20240831, I am thinking about specific problems that can be solved in places where money is already in motion. Although it’s a bit of a leap from the original question, I’ll make a note of one idea that came to mind.

20240903

• Various things come to mind that I want to do:
  • Doing something like Hypothetical Document Embeddings or Helpfeel
  • Playing around with scanning physical books
• However, considering the discussion from nztm 20240831, it’s probably necessary to focus on specific problems at the same time.
• In essence, the only thing that should be done is to create a “genius interface” personally.
  • Since I have PDFs of books that I normally stack up, I will explore ways to make effective use of them for myself.
    • Policy of creating things to use for myself
  • If I can create a useful “genius interface,” then I can think about business again.

=== 20240902 Summary by (claude):

This proposal unfolds an idea for a service that focuses on the inefficiency of book purchases and knowledge acquisition. Currently, many people buy books but lack the time to actually read them, leading to a situation known as “tsundoku.” At the root of this problem is a tendency to make decisions based solely on monetary cost at the time of purchase, without considering the time cost. The proposed service offers an innovative solution to this issue. When users send the physical books they have purchased to a specific address, the service will cut and scan them to digitize the content. Through some kind of “genius interface,” users can efficiently acquire digitized knowledge without having to read the original text. This eliminates the limitations of traditional Kindles and physical books, opening up new possibilities for utilizing the vast knowledge within books. A distinctive feature of this service is the envisioned usage where users continuously send books they are interested in to the service, nurturing their own AI Agent. This has the potential to significantly transform the individual knowledge acquisition process.

(blu3mo)

• https://x.com/bd_gfngfn/status/1830537533236896239

• It’s too understandable that people hoard books because they think books are practically free (the cost in terms of time is overwhelmingly greater than the cost in terms of money).

• I think the structure that leads to “tsundoku” is as mentioned above (blu3mo)
  • Time is the bottleneck
  • Moreover, irrational decisions are made at the time of purchase without considering this, resulting in wasted money (from the individual’s perspective).
• I want to connect books with LLM, but it’s quite cumbersome at the moment.
  • The content of a Kindle cannot be extracted
  • While data can be extracted by cutting, scanning, and OCR of physical books, the pipeline beyond that is not well established as far as I know.
  • By the way, I recently received something like this from BookScan (nishio)
    • /nishio/BookScan AI Summary
    • The quality was terrible, and the minimum charge was 10,000, so I’m a little angry haha
      • I was too busy and forgot that I was angry; I should send an angry message to support haha
• Based on this, I thought of such a service (blu3mo)
  • Users purchase physical books on Amazon and send them to a specific address
  • The service cuts and scans them to digitize the content
    • Similar to https://scanb.jp/
  • Users can make digitized knowledge their own through some kind of genius interface without reading the original text
    • For example: Fractal Summarization
    • For example: LLM presenting information related to content written on Scrapbox in a good way
    • For example: Mapping the content of a book to a VR space
      • (I wrote this as an example of the breadth of possibilities, although a vision of this working well is not clear)
• Key point:
  • Until now, no matter how much a “genius interface” there was, the limitations of Kindles and physical books were the bottleneck, making it impossible to solve.- With this mechanism, one can make a vast amount of knowledge stored in books their own as long as they have a “genius interface.”
• I enjoy exploring and developing a genius interface, so I can do it (blu3mo).
• It is expected that people will continuously send interesting books to this service’s address without hesitation (blu3mo).
• It seems like you could view it as continuously feeding “books” as bait to your own AI agent (blu3mo).
  • This is interesting (blu3mo) (blu3mo).

Things of interest (blu3mo):

• Market size/potential number of users
  • At least I want it (or have the desire for it) (blu3mo).
• Is it legally permissible?
  • https://scanb.jp/tyosakken

  • For customers using Scanpy, except in cases where book data is used for information analysis and machine learning, or when the ordering party holds the copyright, it is necessary to obtain the copyright owner’s work permit for all requested books.

  • It seems like the service is based on the premise that users have confirmed with the copyright holder (blu3mo).
  • Is it an exception when using book data for information analysis and machine learning? (blu3mo)
    • This is concerning.
    • If the original text is processed by LLM without human reading, is it unnecessary to obtain the copyright owner’s permission?
• Can a “genius interface” be created?
  • There seem to be issues mentioned in /nishio/AI要約, but I think this is just an interface problem.
    • It can probably be solved.