[HN Gopher] LLM abstraction levels inspired by fish eye lens
       ___________________________________________________________________
        
       LLM abstraction levels inspired by fish eye lens
        
       Author : swyx
       Score  : 241 points
       Date   : 2024-12-03 16:55 UTC (7 days ago)
        
 (HTM) web link (wattenberger.com)
 (TXT) w3m dump (wattenberger.com)
        
       | Terretta wrote:
       | My cursor got schooled.
       | 
       | Intriguing example with the shoemaker story, but initially
       | confusing. After playing with it, okay, but still perhaps more
       | confusing than the concept which is clear.
       | 
       | The trick with the cursor suggests: what if the context was
       | ambient, like a graph of concept bubbles around whatever
       | paragraph has focus? Or, a step further, as if the paragraph was
       | a card in an array of cards swiped across a desk in "is this your
       | card?" style, with fisheye focus on paragraph text, past and
       | future paragraphs as cards to left and right of this one, with
       | summarized concepts bracketing those from above and below?
       | 
       | In any case, this is onto something, and worth experimenting with
       | various UIs.
        
       | toisanji wrote:
       | Another article with a similar concept, around AI, but not LLMs:
       | https://blog.jtoy.net/choosing-the-appropriate-abstraction-l...
        
       | BitterCritter wrote:
       | I think the process of learning is in fact making larger pictures
       | into smaller quotes and this does not help learning in the way
       | that it's trying to. The process of diluting context is something
       | that fundamentally must happen inside a person and trying to
       | replace that with Ai is contrary to the point.
       | 
       | I do find significant use for this in something like review and
       | considering the legibility of my own note taking, that is one
       | handsome solution.
        
       | baxtr wrote:
       | This sounds uncannily similar to Barbara Minto's "Pyramid
       | Principle"...
        
       | imranq wrote:
       | I think this is a great way to explain the idea of a knowledge
       | graph. There's probably a whole field of ux research opening up
       | to interact with knowledge graphs and using llms to create them,
       | parse them and display them to users
        
       | richardw wrote:
       | A term I've heard since the ...2000's?...is "focus+context" [1],
       | often used in visualisation. I'm looking for a very similar UI
       | construct right now, for balancing simple high-level terms for
       | non-technical people but digging in for people who want more
       | detail, or who want a definition. My thinking is to have a
       | sidebar with more info, but a closable popup might work better
       | for mobile. Don't like popups but it's e.g. how iOS does "look
       | up" when focusing on a word.
       | 
       | [1] And article hints at it with the lines "A fish eye lens
       | doesn't ask us to choose between focus and context" and "This
       | concept isn't new--it's foundational to fields like data
       | visualization."
        
       | ionwake wrote:
       | I have no idea who the author is but that was the most beautiful
       | blog Post I've ever come across. I would be perturbed if didn't
       | know ai was used to help make the fish but I'm not trying to
       | detract from my praise. Good post op
        
         | ruzig wrote:
         | The blog post itself is at product production level.
        
       | aanet wrote:
       | What is this voodoo?? This is the among the loveliest sites I've
       | read in a long time! <3
       | 
       | Beautifully designed, executed and presented. With just enough
       | tension to pull the reader in...
       | 
       | I had to look up the author (who I was not familiar with, tbh)
       | and her other work. Good job!
       | 
       | Beautifully done. 12/10, would read again!
        
         | BoxOfRain wrote:
         | I love the way the fish move around! It's a very pretty webpage
         | in my opinion.
        
       | tianshuo wrote:
       | Great explanation! It's actually a kind of version of "TL;DR,"
       | which stands for "Too long, don't read." One of AI's benefits is
       | personalization, which means that you can jump between
       | abstraction levels, or in other words, information density
       | levels.
        
       | brainless wrote:
       | This is such a lovely article and I relate to the idea as I have
       | thought of something similar when I was building a passion
       | project (https://github.com/pixlie/gitplay).
       | 
       | The idea was to use the Git commit log to show how a software
       | evolved into what it is today. I needed to think of zooming in
       | and out of context, where there are thousands of commits
       | involved. I am not a good UI person so I will definitely take
       | inspiration from this.
        
         | djmips wrote:
         | Very nice! On Perforce I always loved the time-lapse view.
         | Basically another view that could let you see your code evolve
         | over time with branches and changes. There are some git tools
         | like this but never captured my imagination like the Perforce
         | time-lapse. I am eagerly looking forward to trying your tool!
        
           | brainless wrote:
           | Thanks, I just searched for Perforce time-lapse. I will find
           | out more about it, more food for thought.
        
         | newusertoday wrote:
         | nice project.
        
       | p1necone wrote:
       | Oh dear, someone has invented an even more compelling way for me
       | to get stuck in a wikipedia article reading hole.
        
       | dylan604 wrote:
       | we've had other expressions that covered this concept as well:
       | 30,000' view, boots on the ground, deep in the trenches. Forest
       | through the trees.
       | 
       | It's great that someone has a new saying, but it's still the same
       | concept. It's not like we've just been shown a new wheel. It's
       | just been cleaned up and given fresh coat of armor all, and maybe
       | chrome rims instead of the ones from the factory
        
       | sedatk wrote:
       | Isn't this the original design of hypertext as intended by Ted
       | Nelson? Like, instead of clicking links and navigating to
       | somewhere else, the relevant information would actually accompany
       | the text?
        
         | ruzig wrote:
         | I feel the same. The hypertext in this article is AI-generated.
        
       | casey2 wrote:
       | Unless someone starts paying people to read a system like this
       | will probably be inevitable in our information saturated age were
       | nobody has the time to read anymore. Pulling in decades or
       | centuries worth of context and motivation for the passage you are
       | reading is definitely a worthwhile goal.
       | 
       | I really like the layout of the reimagined wikipedia it has the
       | feeling that there is an entire group of people and subjects the
       | care about and rely on the central topic. It's the exact same
       | feeling as studying at a good school. The narrative that this
       | circle creates can be a powerful motivator for learning. With
       | wikipedia and other wikis, most articles feel like your reading
       | an expense report and 99% of links are equivalent to the "general
       | topics" and seem to serve more as distractions from your topic
       | rather than a circle around it.
        
         | kubb wrote:
         | Ever heard of executive summaries?
        
         | TeMPOraL wrote:
         | Part of me wants to say, this is waste of space. The
         | "Reimagined Wikipedia" is basically just taking all the links
         | and rendering them as tiny thumbnails orbiting the article.
         | This adds no new information, no new connections, to what was
         | already present in the article.
         | 
         |  _And yet_ , it somehow _feels_ different. It 's really as you
         | say, "it has the feeling that there is an entire group of
         | people and subjects the care about and rely on the central
         | topic"; those thumbnails beckon, invite you to explore in a way
         | that regular hyperlinks don't.
         | 
         | (That, or maybe after years of getting stuck on Wikipedia,
         | recursively reading linked pages for hours on end, I eventually
         | grew desensitized to regular links.)
        
       | th0ma5 wrote:
       | The biggest problem with complexity is that you have it. I
       | applaud wanting to manage complexity, but there's a difference
       | between managing complexity and simply hiding it.
       | 
       | You want the right information, summarized correctly, from the
       | right source, to the right person, and just in time. We can
       | barely, sometimes, do that if we're lucky with deterministic
       | systems... We loose most of the control of any of these pieces in
       | probabilistic systems. Maybe that works for you, but it feels
       | like hubris to use it anywhere I might actually think I could use
       | it.
        
       | brightmood wrote:
       | How do I disable the fish following my mouse? I skipped the
       | content because of that. Too, damn, distracting.
        
         | danieldk wrote:
         | They don't follow the mouse pointer when using arrow keys
         | scrolling the page (Vivaldi, macOS).
        
         | thirdacc wrote:
         | Bottom right corner, click the "Please, no fish" button.
        
       | acenturyandabit wrote:
       | An outliner would be a good way to represent this info: something
       | like Workflowy, Org mode, Ravel or Dynalist. You get nested
       | bullet points that you can collapse.
       | 
       | Workflowy: https://workflowy.com/ Ravel:
       | https://ravel.acenturyandabit.xyz/
        
       | georgestrakhov wrote:
       | https://www.telescopictext.org/
        
         | TeMPOraL wrote:
         | Yes, like that, but less interactive. Clicking to expand every
         | part individually is tedious, especially with no up-front idea
         | of how many levels of detail are there. TFA's example operates
         | on lager blocks of text, too, which seems to work out better.
        
       | hopfog wrote:
       | Beautiful blog post!
       | 
       | I've built something similar and have yet to find a compelling
       | use case for it. In my mind it seemed like such an obvious idea
       | that I thought I'd start using it instead of Wikipedia. However,
       | in practice I seldom care about the broader or more specific
       | topics for the thing I'm looking into. It's strange though
       | because I'm a compulsive rabbit holer but it just doesn't trigger
       | the same urge to explore related nodes as Wikipedia does.
       | 
       | With that said, this UI is much slicker so maybe that is the
       | missing piece.
        
       | internet_points wrote:
       | As long as I can turn it off, I get distracted so easily. I can
       | barely code full-screen with one function visible at a time.
       | 
       | That wikipedia view sounds terrifying.
        
       | stavros wrote:
       | I created something related long ago, when there weren't any
       | LLMs. I do think it works better when human authors can choose
       | when to add information and how:
       | 
       | https://skorokithakis.github.io/expounder/
        
       | bux93 wrote:
       | Those who don't know the pyramid principle are doomed to reinvent
       | it.
        
       | nyanpasu64 wrote:
       | Haystack Editor uses a UI of inter-function control flow graphs,
       | but lacks the ability to "summarize" surrounding code. (In my
       | usage I found it was buggy, and useful for tracing a specific
       | level of knotty code, but had issues relative to traditional IDE
       | tabs/panes for both small files and flipping between multiple
       | "aspects" of a system.)
        
       | tubs wrote:
       | The example is incorrect. On the paragraph where the elves
       | (dwarves??) get the clothes, it is specified no cut pieces are
       | left out for work, but the subsequent summary is the elves settle
       | to work?
       | 
       | I don't think this is a particularly good showcase.
        
       ___________________________________________________________________
       (page generated 2024-12-10 23:01 UTC)