Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Building a Knowledge System That Enhances Rather Than Replaces Thought

nsavage.substack.com

Post reply on HN
Print view View original

21–30 of 54 posts

Re: Building a Knowledge System That Enhances Rather Than Replaces Thought

#21

Post by lukev »

nsavage wrote: post #20
visarga wrote: post #14

> When you're dealing with ten notes, it's trivial to dump them into an LLM and generate connections. With a hundred notes, it's still manageable. But what happens when you hit 10,000 notes? Or a million? We quickly run into the limitations of context windows and processing capacity. Vector search helps narrow things down, but it's prone to missing important connections that a human mind might naturally make. I run i…

Author here. Glad I'm not the only one with this problem! What I'm experimenting with is building a graph out of "entities", which I'm calling people, places and things found in each card. The point I guess is to allow the LLMs to make connections between cards, but this won't be replacing user made connections. This lets me pick up all cards that reference something like, I don't know, "factories" as a concept, whic…

I have a thesis that RDF is good for this sort of work (I gave a talk on it at Clojure Conj last year, https://youtu.be/OxzUjpihIH4 if you're interested.)

It's good for bridging the gaps between LLMs, ontologies, traditional structured data, and logic programming.

Re: Building a Knowledge System That Enhances Rather Than Replaces Thought

#22

Post by mh- »

RaftPeople wrote: post #8
vouaobrasil wrote: post #5

> What parts of note-taking should we digitize? What aspects should remain firmly in human hands? And most importantly, how do I create a tool that enhances rather than replaces human thought? My personal philosophy is to use the most primitive methods possible and only use technology when there really is a strong need to go to the next level. It exposes what I really need, what are the weaknesses, etc. For example,…

> Moreover, writing things first by hand helps me remember them better and "feel" the knowledge through my hands. There was a science article recently that studied taking notes and it's results were that handwritten notes improved recall compared to typed notes.

It cramps my hand pretty badly to handwrite the amount of notes I prefer to take in live meetings, so I really dislike that this is true. (And anecdotally, it is true for me.)

Re: Building a Knowledge System That Enhances Rather Than Replaces Thought

#23

Post by ketedrum »

lukev wrote: post #21
nsavage wrote: post #20

Earlier quoted context omitted.

Author here. Glad I'm not the only one with this problem! What I'm experimenting with is building a graph out of "entities", which I'm calling people, places and things found in each card. The point I guess is to allow the LLMs to make connections between cards, but this won't be replacing user made connections. This lets me pick up all cards that reference something like, I don't know, "factories" as a concept, whic…

I have a thesis that RDF is good for this sort of work (I gave a talk on it at Clojure Conj last year, https://youtu.be/OxzUjpihIH4 if you're interested.) It's good for bridging the gaps between LLMs, ontologies, traditional structured data, and logic programming.

Interesting, thank you! Gonna watch this asap

Re: Building a Knowledge System That Enhances Rather Than Replaces Thought

#24

Post by visarga »

austinjp wrote: post #16
visarga wrote: post #14

> When you're dealing with ten notes, it's trivial to dump them into an LLM and generate connections. With a hundred notes, it's still manageable. But what happens when you hit 10,000 notes? Or a million? We quickly run into the limitations of context windows and processing capacity. Vector search helps narrow things down, but it's prone to missing important connections that a human mind might naturally make. I run i…

Intriguing. I've been thinking of something similar. Do you have any notion of what constitutes an 'idea' in this context?

I am using a LLM to extract ideas from text, as a numbered list, each one self contained, about 50-100 words long. It also generates relations between ideas (supporting, contradicting, expands on, etc...)

The idea extraction step is simple when starting from scratch, but when you got megabytes of text deduplication becomes a problem. I like to circle back a lot.

Re: Building a Knowledge System That Enhances Rather Than Replaces Thought

#25

Post by uoaei »

lukev wrote: post #21
nsavage wrote: post #20

Earlier quoted context omitted.

Author here. Glad I'm not the only one with this problem! What I'm experimenting with is building a graph out of "entities", which I'm calling people, places and things found in each card. The point I guess is to allow the LLMs to make connections between cards, but this won't be replacing user made connections. This lets me pick up all cards that reference something like, I don't know, "factories" as a concept, whic…

I have a thesis that RDF is good for this sort of work (I gave a talk on it at Clojure Conj last year, https://youtu.be/OxzUjpihIH4 if you're interested.) It's good for bridging the gaps between LLMs, ontologies, traditional structured data, and logic programming.

Do you know of any good RDF datasets or repositories to look for? I am familiar with linked data / semantic web but have not kept up on developments for a long while now.

Re: Building a Knowledge System That Enhances Rather Than Replaces Thought

#26

Post by JellyBeanThief »

voidhorse wrote: post #15

After trying out a bunch of digital zettelkasten tools, I just went back to paper. This take may be a bit hot, but I actually think paper and pen is already the optimal maxima for thinking —not for retrieval, mind you, but for helping us produce new thoughts . Zettelkasten (at least the way Luhmann used it) is meant for this purpose—it is not a system for storing information and retrieving it but rather for supportin…

> the ability to organize things freely in space

This is the key. It's not enough for digital tools to just put things in folders or tag them. It's the links themselves that need elevation. People need to add metadata to links, they need to apply rules specifying what links to crawl and how to arrange them in space, they need to specify how content should be displayed in islands of connected content. Then they need to be able to arrange islands on a 2d, 2.5d, or 3d canvas.

We have information input and retrieval solved. For some reason it's taking a real long time for people to get to spread out.

Re: Building a Knowledge System That Enhances Rather Than Replaces Thought

#27

Post by exe34 »

harlanji wrote: post #17
voidhorse wrote: post #15

After trying out a bunch of digital zettelkasten tools, I just went back to paper. This take may be a bit hot, but I actually think paper and pen is already the optimal maxima for thinking —not for retrieval, mind you, but for helping us produce new thoughts . Zettelkasten (at least the way Luhmann used it) is meant for this purpose—it is not a system for storing information and retrieving it but rather for supportin…

Tried Inkscape? I was all paper for a long time but it's taken over as I've gotten used to it an the hot keys. I carry a little Wacom tablet that works good with it too, but seldom even get it out of my bag. It does have a (configurable) fixed page size, no ability to multi-page which was kind of annoying but I started thinking of it as a feature as well.

have you come across scribus?

https://impagina.org/features

you don't have to use all the desktop publishing features, but it does have multi page, flows the text from page to page, some basic vector drawing etc.

Re: Building a Knowledge System That Enhances Rather Than Replaces Thought

#28

Post by vouaobrasil »

mh- wrote: post #22
RaftPeople wrote: post #8

Earlier quoted context omitted.

> Moreover, writing things first by hand helps me remember them better and "feel" the knowledge through my hands. There was a science article recently that studied taking notes and it's results were that handwritten notes improved recall compared to typed notes.

It cramps my hand pretty badly to handwrite the amount of notes I prefer to take in live meetings, so I really dislike that this is true. (And anecdotally, it is true for me.)

I used to write a lot and one thing that helped was switch to a mechanical pencil and use 2B lead. It's very soft and requires very little pressure, even compared to pens.

Re: Building a Knowledge System That Enhances Rather Than Replaces Thought

#29

Post by wizzard0 »

vouaobrasil wrote: post #28
mh- wrote: post #22

Earlier quoted context omitted.

It cramps my hand pretty badly to handwrite the amount of notes I prefer to take in live meetings, so I really dislike that this is true. (And anecdotally, it is true for me.)

I used to write a lot and one thing that helped was switch to a mechanical pencil and use 2B lead. It's very soft and requires very little pressure, even compared to pens.

I had tried a ton of options, and settled on ceramic ball pens.

They're both the softest and most robust on different kinds of paper in case you don't have your preferred notebook/whatever handy

Re: Building a Knowledge System That Enhances Rather Than Replaces Thought

#30

Post by dflock »

harlanji wrote: post #17
voidhorse wrote: post #15

After trying out a bunch of digital zettelkasten tools, I just went back to paper. This take may be a bit hot, but I actually think paper and pen is already the optimal maxima for thinking —not for retrieval, mind you, but for helping us produce new thoughts . Zettelkasten (at least the way Luhmann used it) is meant for this purpose—it is not a system for storing information and retrieving it but rather for supportin…

Tried Inkscape? I was all paper for a long time but it's taken over as I've gotten used to it an the hot keys. I carry a little Wacom tablet that works good with it too, but seldom even get it out of my bag. It does have a (configurable) fixed page size, no ability to multi-page which was kind of annoying but I started thinking of it as a feature as well.

Inkscape does have multi page, now, if you still want that.
Post reply on HN