Now that I’ve succeeded in making non-public materials the target of vector search, it’s time for Cross-sectional Vector Search, but at this time, I’m having trouble with the unit of the chunk of information…

  • Now, if you’re targeting Scrapbox Cross Search, you’ve checked the box.
    • I was thinking, ā€œI’ll make it so you can switch search targets later.ā€
    • Not yet implemented.
  • Books?
    • One book is one unit, that’s a thousand checkboxes, okay?
    • Bundling?
      • Bundling by the same author is the same atmosphere as Scrapbox
      • Audrey Tan’s book, with mixed authorship, bundle?
        • You can’t put it all together mechanically by the author, can you?
      • It’s impossible to make proper groupings for a thousand books in advance.
    • All in one?
      • If you consider that a page in a book is a page in Scrapbox, think of 300 pages and 1000 books, and suddenly about 20 times more than my Scrapbox will appear!
        • dangerous-looking
    • Oh no, it doesn’t have to be all books.
      • You could add a few ā€œbooks I have a lot about the same personā€ in clusters, or you could add a few more.
      • Virtual ā€œScrapboxā€ image of the person
      • More than a page in a book, more than a page in Scrapbox…
        • I felt like the publisher would get mad at me for trying to say ā€œpoor quality.ā€
        • Since we have different definitions of ā€œquality,ā€ I’ll come up with another way to say it.
        • A page of a book is basically just a long, sticky sentence that is mechanically cut to the same size, so there are no ā€œshort, crystallized pagesā€ like there are in Scrapbox.
        • I think this will probably bring ā€œpoor quality search resultsā€ to destinations that do vector searches.
  • Not to be the largest from the beginning.
    • Possibly something we can see in the process of increasing scale.
    • If you start out with the largest scale, you lose the opportunity to see that.
    • And even the current Scrapbox cross-vector search is narrowing people down to begin with.
  • Is the checkbox necessary in the first place?
    • A checkbox can represent 2^N pieces of information, but do we need more than ā€œall acrossā€ and ā€œone specificā€ to begin with?
    • Ah, so that’s where you get to ā€œWhen books are grouped by book, I want to cross-search from books by the same authorā€.

@miiura: The API update has dramatically reduced the execution time of the agent. This is a significant impact for autonomous agents. šŸ¤–

I want to tinker with this one too, not enough time!

  • For careful deliberation, sharing of prerequisite knowledge is necessary. However, people cannot learn what they do not want to know. Therefore, for deliberation, it is first necessary to arouse ā€œthe desire to knowā€ in a large number of people.

  • Why not visualize the lack of learning and ask, ā€œDo you want to learn or leave the decision to those who have learned?ā€ --- 648ad4c7aff09e000019926a

    • @_kayato: Even as social issues become more complex and fragmented, people’s cognitive capacity is limited. I feel that deliberation based on prerequisite knowledge has become something that only a few people with the luxury of time can participate in. If it is based on curiosity, there is also the problem of the filter bubble, which is a very difficult issue…

      • @nishio: sure…I wonder if we could use LLM to streamline learning of prerequisite knowledge and remove bias…

  • Diary 2023-06-14 ←Diary 2023-06-15 → Diary 2023-06-16 100 days ago Diary 2023-03-07. 1 year ago Diary 2022-06-15.


This page is auto-translated from /nishio/ę—„čØ˜2023-06-15 using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I’m very happy to spread my thought to non-Japanese readers.