2024-05-01

  • System for one book

    • Try it in the Plurality book
    • temporal context.
    • The Intellectual Production of Engineers
      • The Japanese version cannot be published as is.
      • There is an English version.
  • System for one Scrapbox

  • System for multiple books

    • Fun to discover links between different authorsā€™ descriptions
    • We have data.
    • Cannot be made public.
    • No, wait, thereā€™s a way: Gutenberg.
  • Multi-person Scrapbox projects

    • If you are co-editing
    • When referring to a link
  • Fractal Summary

  • Assuming a one-dimensional structure.

  • If you ignore cost, you can do a fractal summary of a single book.

    • I hope you understand the structure of the chapters and headings, though.
    • As for Plurality books and your own books, you can add metadata manually.

[/blu3mo-public/Fractal Summarizer feedback summary](https://scrapbox.io/blu3mo-public/Fractal Summarizer feedback summary).

  • 2024/4/24
  • I was just doing something else and it occurred to me that if the terminal input is ā€œpeopleā€™s diverse opinionsā€, it could be considered a form of broad listening.

    • Right now itā€™s chunking based on the order of input, so if you input a variety of opinions in different orders, youā€™re going to get a fractal summary that you donā€™t understand.blu3mo.icon
      • I was imagining it as an embedded vector and clustering it with different kā€™s using the k-means method.nishio.icon
        • I see blu3mo.iconblu3mo.icon.
    • If we can chunk based on meaning rather than order, we can generate summaries that help us understand the diversity of opinions.
nishio.icon 2024-05-01 - [[KJ method]] collects related labels and then bundles them together with a nameplate - This can be seen as a summary - [[KJ legal effect of repeatedly applying summaries]]. - from [[Current impressions and thoughts on OMNI]] - 'Related' can be 'Scrapbox links' or 'vector similarity.' - Now that's just the "back and forth" one-dimensional fractal summary. - Where to cut a one-dimensional series of graphs - Currently, I think the size (number of pieces) is determined first, and then the pieces are cut. - [[Vector Search in Nishio]] also determines the number of tokens and cuts them by that number. - Commonly used rustic methods - I'd better see what it means and chop it up. - [[MemoChat: Tuning LLMs to Use Memos for Consistent Long-Range Open-Domain Conversation]]

Iā€™m also thinking it would be interesting to create something that bridges national languages in Japan, Korea, and Taiwan with Face the Ocean, but it doesnā€™t seem to improve my intellectual productivity.

The multilingual glossary in the Plurality book is also a hassle when I think about the hassle of connecting to the original website.


This page is auto-translated from /nishio/ę€č€ƒć®ēµēƀē‚¹2024-05-01 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. Iā€™m very happy to spread my thought to non-Japanese readers.