[HN Gopher] Building a Knowledge System That Enhances Rather Tha...
___________________________________________________________________
Building a Knowledge System That Enhances Rather Than Replaces
Thought
Author : nsavage
Score : 146 points
Date : 2025-01-02 18:47 UTC (1 days ago)
(HTM) web link (nsavage.substack.com)
(TXT) w3m dump (nsavage.substack.com)
| uludag wrote:
| This article touches on a tension I've been feeling too with the
| rise of LLMs. I feel that analogies can be drawn to time
| management and systems of productivity, like GTD. Like in a GTD
| system, you almost need the various projects, tasks, next steps,
| deadlines, commitments, etc. to live in your head so you can make
| intuitive decisions about what you should be working on at the
| moment; something that AI almost by definition can't do, since
| your subjectivity is essential in this process, but in certain
| ways may be able to assist.
| treetalker wrote:
| I've been sensing this too: people seem to forget about the
| principal and beneficial internal effect on the mind that
| results from mental exertion. I raised a similar point in
| another HN post today and someone downvoted me because they
| interpreted my statement as being too anti-LLM!
|
| We're reliving Socrates's lament about writing.
| nuancebydefault wrote:
| Gave you an upvote to compensate
| Terr_ wrote:
| > From Socrates worrying that writing would destroy memory [...]
| Think of it as a partnership: the computer handles the
| organizational heavy lifting, while you focus on the thinking.
|
| I'm less worried about memory _per se_ and more about failing to
| think, or getting brainwashed /ring-led by a system with its own
| biases and quirks. Any sufficiently complex organizing _is_
| thinking!
|
| Perhaps the simplest example is when quantities (numbers, easy to
| record) get a mental weight that overshadows and hides their
| dimensions (the definitions, what they really mean.) For example,
| a tendency to automatically assume a rising GDP number is an
| unqualified good sign.
|
| Stuff like LLMs bring that into newer and more-dangerous
| territory, because the model also contains uncountable subtle
| biases from its training data, and even if you _know_ it isn 't
| aligned (heh) with your own mental models you can't reliably
| change it. Much like false-memories implanted by interrogators,
| patterns in those systems can and will leak into the users.
| Whenever we can't "think about how our thinking is being
| changed", I'd say that's _axiomatically bad_.
| Terr_ wrote:
| P.S.: Lest anyone think I'm a Luddite--not that I think that
| appellation is actually that bad--an example where I _would_
| use an LLM would be to help me generate synonyms or alternate
| inputs to a more-traditional search for a discrete external
| piece of information.
|
| For example, I might remember a book with a jester playing a
| lute and singing about ogres, and I just can't find any clear
| search results, because it was actually a bard strumming a harp
| with a poem about giants.
|
| This is much less dangerous than just throwing every dang thing
| into LLM inputs (since prompt injection isn't a fluke, it's a
| way of life) or filtering the results back through the same
| model in an opaque fashion.
| vouaobrasil wrote:
| > 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, I take all my notes with
| pen and paper. But if I find that I'm really referring back to
| something, I might write it up in a document. I don't see the
| point in digitizing everything right away if I'm never going to
| use what I write.
|
| Moreover, writing things first by hand helps me remember them
| better and "feel" the knowledge through my hands.
|
| Same thing with photography. I don't tend to use the burst mode
| on my camera unless I REALLY need it. When it comes to
| accomplishing things, I found (personally) that asceticism with
| tools is best.
| RaftPeople wrote:
| > _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.
| mh- wrote:
| 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.)
| vouaobrasil wrote:
| 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.
| wizzard0 wrote:
| 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
| hunter-gatherer wrote:
| > 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.
|
| This is a good take on life in general. My parents were atmy
| house recently and my mother commented on how we never have
| dirty dishes. I told them the solution to me was always
| obviouse; don't own a dish washer. My wife and I have never
| owned a dishwasher since we were married, which has forced us
| to be more discriminatory about our dishware & forces us to
| wash dishes immediately after use so they're available again.
| People with dishwashers tend to have double what they need
| because they wait to start a load until the washer is full.
| Lots of the available technology we have has progressed past
| the level of optimization IMO.
| Kalq wrote:
| With regards to zettelkasten, I've always worndered where
| serendipitous discovery of notes goes from being actually
| worthwhile to frequently getting distracted but justifying it as
| productive.
| randomcatuser wrote:
| > create something that thoughtfully augments human intelligence
| rather than replacing it
|
| I htink this is a really cool design space, particularly with the
| framing of the _purpose_ of Zettelkasten as reflecting your
| personal understanding of the world, your views, rather than a
| cold, dead categorization of facts and objects.
|
| How can we design software systems that enhance our sensemaking?
|
| - Encouraging us to ask more questions, or write down our
| thoughts? You can imagine the AI bringing snippets to your
| attention, asking: "What do you make of this?" or "How might we
| connect these?"
|
| - Augmenting our notes? Finding nubs of thought, and presenting
| research/articles to help us develop those lines some more
|
| - Collaboration suggestions? People who wrote about the same
| things?
| gmuslera wrote:
| Try to build bridges, not roads. Sometimes your flow is
| interrupted or is inefficient because you are searching for the
| right word, or have a small understanding of a concept. or a
| better way to express a concept. Try to have a tool that does
| that small push. but without overwhelming you with data and
| disrupt your own process.
|
| You usually are not writing for someone else, but for you, and
| the reward is the process you went through more than the final
| text. Maybe the user want to go with an AI all the way in, or
| not, but it must be user decision,and by default it should be
| something small and not intrusive.
| le-mark wrote:
| Recently I've been imaging a world where social media algorithms
| were tuned to help people instead of "driving engagement" with
| ever more outrage bait. Oh you're watching clips about machining
| and by your data profile you're an uneducated adult? Here are
| some trade school, financial assistance, and self help links to
| nudge you toward a better life! What a world that would be.
| BadHumans wrote:
| Isn't this what Tiktok is in China? It mostly promotes
| educational and science content.
| drekipus wrote:
| yes, for natives.
|
| but it also ties in with censorship/propaganda, ie: "no such
| thing as homeless people in china"
|
| If you're a westerner, it will show how bad US is, lusty
| images, drug use, gambling. etc.
|
| It's a super double-edged sword. We could have it in the west
| but we'd need a government that is looking out for the
| interests of people, not
| israeli/multinationals/lobbies/industry etc.
| henning wrote:
| it's hard to understand what problem this person wants to solve.
|
| is a version controlled directory of Markdown files + grep
| enough? or a super trivial wiki Rails app that will turn WikiText
| into links that you could probably prototype in an hour or less?
| why or why not?
| UmYeahNo wrote:
| What you're describing sounds an awful lot like a barebones
| obsidian vault.(https://obsdian.md)
|
| I'm not familiar with the OP's notion of a "zettlegarden" but I
| get the impression that they want something a bit more than
| just markdown files. Perhaps the idea is to increase the notion
| of connection between captured notes, whether intentional
| created, or "serendipitous".
| visarga wrote:
| > 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 into the same problem. I made a RAG system and imported 15
| years of reddit and HN content (my own message logs) + 2 years of
| LLM chats, totaling about 80MB of text. I can use it to retrieve
| fragments but there is duplication, and almost all duplicate
| fragments have slightly different approaches. How do I merge all
| of them, and how do I get a deduplicated taxonomy? I got about
| 290K unique keywords extracted from the text, it doesn't fit into
| LLMs.
|
| I am gravitating towards building graphs of ideas, and having a
| way to generate unique (non duplicate) new ideas while ingesting
| new text.
| austinjp wrote:
| Intriguing. I've been thinking of something similar. Do you
| have any notion of what constitutes an 'idea' in this context?
| visarga wrote:
| 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.
| nsavage wrote:
| 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, which isn't something the
| user will necessarily want to directly connect.
|
| Downside is that deduplicated taxonomy: I have, from a scan
| through my list, "factories", "Factory" and "Factories". My
| attempt to head this off is by vector searching based on the
| proposed entity keywords and having an LLM decide if they are
| the same or not. I think I need to work on my prompt for this
| though.
| lukev wrote:
| 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.
| ketedrum wrote:
| Interesting, thank you! Gonna watch this asap
| uoaei wrote:
| 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.
| zozbot234 wrote:
| You can refer to the Linked Open Data (LOD) cloud for a
| network of openly available RDF datasets. Wikidata
| currently serves as the unofficial "hub" for the cloud, a
| role that was formerly played by DBpedia.
| spothedog1 wrote:
| yup I've been working on building a platform entirely built
| on RDF Ontologies. Excited to try to combine it with LLMs
| in the not-to-distant future.
| voidhorse wrote:
| 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 supporting the _creation_ of new connections.
|
| The computer aided tools are suboptimal for this because they
| lack the good constrains of paper (severely limiting the search
| space) and the good features (seamless ability to incorporate a
| variety of representational modes, text, image, equation, with
| zero overhead, and the ability to organize things freely in
| space). As much as the digital knowledge base sounds good in
| theory, I don't think it will ever be as optimal for _generation
| of thought_. If all you want to do is summon existing
| information, digital tools are evidently superior. I personally
| think a hybrid system where one "thinks" on paper and "archives"
| digitally (after the thinking is done) might be best, but
| ultimately, we will be most productive with whatever system we
| actually _enjoy_ using.
| harlanji wrote:
| 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.
| exe34 wrote:
| 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.
| dflock wrote:
| Inkscape does have multi page, now, if you still want that.
| JellyBeanThief wrote:
| > 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_.
| egormakarov wrote:
| I sometimes dream of video stickers, where you could put some
| short video on it, and then just put it on the paper notebook
| page
| buggy6257 wrote:
| Hypothetically with a sticker printer like a Cricut or
| something you could do little QR stickers...
| Dennip wrote:
| would something like an ipad + apple pencil not bridge this
| concept? The usage of pen & paper with the addition of the
| digital world; OCR, tagging etc.
| sdwr wrote:
| You would think so, but the micro-interruptions of switching
| modes, selecting documents, selecting tools, etc make a huge
| difference in terms of flow.
|
| It sounds small, but "taking a second to select a new
| brushstroke from a list of options" is exactly what OP is
| talking about in terms of raw creation vs synthesis.
| Dennip wrote:
| I guess so, but those are also all optional right? IF you
| would only write freehand with a ballpoint in real life
| then you wont use any of those menu options on-device
| either? IIRC the native notes app on iPad will dynamically
| create bullet lists, rule lines etc from your freehand
| typing without needing to select things.
| sdwr wrote:
| Every piece of the digital experience is (in this case)
| simulating the real world imperfectly. Paper is, screens
| are pixels and direct light and refresh rates. A pen is,
| a menu option is not.
|
| There are definitely advantages to digital, but there's
| no denying that, by adding a middleman between yourself
| and the world, you lose raw experience.
| 1123581321 wrote:
| There are a couple apps that try to address that problem.
| My favorite is Muse. You just open it up, pan to a blank
| area and start writing. Tools are deliberately limited so
| there's no point futzing with colors and thicknesses.
| There's no document selection, just optional nested boards,
| so your spatial orientation is always engaged, like with
| paper.
| LeroyRaz wrote:
| Why do you prefer paper to a word processor or latex?
| jrm4 wrote:
| Wow, so glad to see the author's idea and this as well.
|
| After playing with this stuff it hit me why these systems are
| simultaneously so attractive and also not amazing.
|
| Human brains are still way better at the very cool linking
| thing. We make those connections better than the bots.
|
| Which is why I dream of, and tried to start work on one that
| was basically "cards."
|
| I'd love to see one with "the card" as the main atom/unit --
| but that had a zillion different visual "arrangement" methods.
| Give me alphabetical, give me semantic, give me random etc etc.
| two_handfuls wrote:
| The Livescribe pen goes well with this philosophy.
| Rochus wrote:
| > _It 's reminiscent of Luhmann's original zettelkasten system,
| where physical proximity created a kind of "neighbor effect" that
| digital systems struggle to replicate._
|
| That's what outliners are good for. You automatically generate
| context and proximity when adding/indenting outline items. And we
| have the thing called hypertext for cross-referencing and
| navigation. I think Luhmann's zettelkasten is a quite poor
| concept in comparison, because the cards are small and don't
| adjust to the size of a topic, so we have to split topics, and we
| have to take care of identification and referencing the cards
| ourselves. When translating zettelkasten to digital, then at
| least we can get rid of some of these limitations, but we still
| don't have all the advantages of outliners.
|
| In my experience, it is purely a matter of habit whether you
| write down ideas or thoughts by hand on paper or enter them
| directly into an outline on the computer. Before I had tools like
| Ecco or my own CrossLine (https://github.com/rochus-
| keller/crossline/), I also produced a lot of notes on paper.
| After a few years of working with these tools, I got rid of paper
| completely. With CrossLine, I also solved the problem of linking
| to other documents or emails, so that I can link any information
| or put it in the necessary context. And it scales well. I have
| outline files with thousands of notes, minutes, concepts,
| specifications and the like, everything cross-referenced and
| full-text searchable. I can still reproduce e.g. meetings or
| paths to decicions taken ten or more years ago with little
| effort.
| pxeger1 wrote:
| I write with a similar approach to zettelkasten. My technique is
| purely digital. Paper is too easy to lose, damage, or not have on
| you when you need it, compared to a digital file that can be
| copied, synced, and backed up.
|
| I keep everything in just one long file, which I think gives a
| similar feeling to what the article describes about notecards -
| by using folds and reordering where necessary it's quite easy to
| always have whatever you want visible; there's nothing forcibly
| hidden from you by a file structure.
|
| I use plain text (with a tiny bit of custom syntax highlighting)
| in vim, because I like the simplicity and it fits my needs. But I
| think a one-file approach could work fine in a Markdown or
| WYSIWYG format.
___________________________________________________________________
(page generated 2025-01-03 23:01 UTC)