This is very good. I can't put my finger on it, but it seems more important than a mere "gimmick." I noticed that if you click on a topic already explored, it won't open again. That's cool, I'd make it snap back to the pane where it's open.
Kudos! This is an interesting perspective on how we really need to put a little more effort into the UX of LLMs.
If "delve" was meant to be an in-joke, I just wanted you to know: I got it.
I also have a Custom GPT "AutoExpert (Chat)" [0] that several reviewers have called "the perfect Rabbit Hole GPT" due to the way it leads users through learning a topic. You might dig it, especially since free tier users have access to these now.
what's the joke?
ChatGPT overuses the word "delve" in its writing. Search your email for "delve," and look at how common it starts becoming (esp. in marketing emails) around the time ChatGPT takes off.
That's because chatGPT talks to hundreds of millions of people and puts a trillion tokens in their heads per month. And out of every 1000 tokens, a "delve" creeps up.
Ah gotcha lol
Oh I like this! Love how easy it is to dig deeper. Worth noting that free users are rather restricted in their uses of custom GPTs so probably won’t be able to dig as deeply as they’d like.
EDIT: I’ve now quizzed it about string theory, quantum mechanics, classic Roman pasta dishes, Italian wines, and sent it a picture of some poison hemlock and I think I’ve found my new favourite GPT. Great work!
EDIT 2: asked it to critique a photo I took recently and that was great too, really impressed with this.
Wow, I find this very useful. The eliciting an expert, alternative experts, oppositional viewpoints... fascinating.
We're just getting started on what we can do with LLMs!
Are you paying for the API calls yourself here?
It seems like they're caching information/replies on key words which is a good optimization.
I assumed it was coming straight from the API because of the token-by-token generation effect but maybe you're right.
The token-by-token responses are probably API, while the "instant" loads seem to be cached.
Probably using groq based on speed of response
I have a lot of OpenAI API credits to use by end of the month, so I’m using 4o. I’ll probably switch to a more sustainable model afterwards, consider this a request for API credits, all! Email’s on website :)
Taking a step back: The UX/UI for LLMs in general are very immature. We're in the very early days of to best interact with these tools. We need more experimentation like this to help figure out what works, and doesn't work.
Kudos!
Hopefully HCI improves enough that interfaces will be a last resort.
there's a chrome extension, "HLAI" that's playing around with this idea, too
I will say the bare bones chat interfaces are so so much better than the awful copilot side panes, and quasi-material designed to death Google attemps at interfaces so far. I am sure with multi-modal, and with special cases for deep research there may be improvements, but insofar as straight text chat is concerned I think the simplest interfaces are hard to improve upon.
This is great and something that I've wished existed. Thanks for making it! Right now, the tiles are linear. E.g. if tile A links to tiles B and C, clicking either B or C will open a new tile directly next to A (and only one of B or C is visible at a time). What do you think about making more of a tree layout where B and C both branch from A and can be viewed simultaneously?
I agree with this. The UI is already great, but a tree-like structure would be awesome.
Seconded (or thirded) -- a way to navigate tree conversations is desperately needed. Perhaps something like what Gingko [1] does.
I think this is a point where it’s helpful to take a step back in scope — instead of looking for LLM tree UI implementations, we should consider the mature field of general text hierarchies. I’m lazy, but I posit there are many, many UIs for visualizing wikilink-esque document repositories, such as obsidian plugins (?), browser extensions, vim/emacs/other plugins, etc.
Personally, I’m a huge fan of a few principles that I hope to impose on the world via book, eventually:
1. Indices over keys.
2. No single set should have more than N elements, where N is usually 10 but could be 2/3/4 if you’re doing decision trees, and could be 16 if you’re insane and want to use hex indices.
3. Each element can be referenced locally with a simple index (`3`), or a full path made by concatenating the indices of its ancestors from the root (`053`).
This would be an example of an “analytic” approach, as opposed to the ad-hoc “synthetic” approach of just visualizing whatever wikilink structure there happens to be. There’s a huge space of solutions “between” these two - such as constraining the ad-hoc visualization to meta-tagged wikilink relations — but I think the dichotomy is useful.
Personally, I prefer to use predesigned structures wherever possible for exactly these reasons. It makes automated visualization possible, in many cases… An example would be reusing the same 3/4 12-element directory template for every SWE project. I hope it’s clear how the same idea could be directly applied to a research project performed with lots of automated LLM queries.
Love it! I like that the site is straight to action, but I think it could really benefit from a walkthrough. Here’s an idea!
It would be great if we had an introduction to the site right in the prompt! to help understand its main purpose right from the start.
It'll be great if the first thing you see is [Explain what "delve into" is] as a prompt suggestion. Next, it will reply with, "It’s for exploring topics deeply, similar to going down "rabbit holes" where one interesting thing leads to another. Here are some examples ..."
Then, you guide the user through the functions step-by-step. Something like, "Click on option X to start a new thread, then choose from the suggested prompts or create your own. Follow the flow to see related threads and dive even deeper."
My 2 cents here is that it’s less obvious that this would be a net positive, people fall into two camps on these type of getting started suggestions. Many will say this very guided walk through is an obvious useful feature, and many will say that it annoys them.
I’ve gathered a lot of feedback on things like this for a few different sites and apps from senior UX designers and PMs who contradict eachother on improvements and best practices all the time and from users. You’ll of course only hear from the people who want it rather then the ones who would be annoyed by it :)
Great project that seemed very easy to understand and straight forward to me, no further walk throughs needed ;)
perhaps consider a tutorial you can close with just one click? those always seem a good compromise
compromise, the father of all mediocre designs
This could replace the hours I spend on Wikipedia. Hope it's not too expensive to run.
Thanks <3
I have some API credits but I intend to make it sustainable. If any LLM provider wants to sponsor some credits, hit me up.
Otherwise I’ll switch models and add user accounts.
It looks like the frontend is a React SPA. What is the backend stack?
If your backend is is javascript and/or depending on how complicated it is, an easy idea might be to allow users to paste their own OpenAI API token in and have it use that. For various security/privacy reasons it would be ideal if the API calls came directly from the frontend in that case though, and given the caching implementation and other things I'm guessing decoupling that might be pretty challenging. Figured I'd throw out the idea anyway though.
But not as accurate. A few quick queries and already it's providing misinformation. As much as Wikipedia has problems with truthiness in some instances, it's not nearly as bad as misinformation-ridden GPTs.
Great idea. I also see inspiration from Andy Matuschak's notes [0], of which I'm a bit fan
[0] https://notes.andymatuschak.org/zB74H9CuWrosEuqve7jZyCo?stac...
Damn, is there some SAAS like notion or an Obsidian plugin to be able to take notes like this?
https://forum.obsidian.md/t/andy-matuschak-mode/70
There's an Obsidian plug-in.
Warning: if you plug this into tvtropes then global productivity will drop sharply.
?
It's a meme, because tvtropes is insanely addictive to dive deeper into the various rabbit holes it has.
If you enable this kind of rabbitholing it'll be even more insanely addictive because of how awesome it feels to explore rabbit holes.
This UI is fantastic! Was the name "delve" AI generated? I just cannot believe a human uses that word...
time for you to delve into more literature.
It's obviously a reference to the meme going around about how ChatGPT uses that word inordinately frequently.
I'd love it if it had a "zoomed out" tree-like view that makes all the different paths of conversation viewable at once
tldraw (make real) is close to this, but on a canvas: https://news.ycombinator.com/item?id=38288130
That's sort of a mind map. We are building/experimenting with something like this . https://iwtlnow.microschools.in/
You can either enter your GPT key, or fill in the form here https://learn.microschools.in/ and we'll give you access if you'd like to give it a spin.
I like it a lot. Feels like idle Wikipedia link surfing but with the key difference that each new step keeps track of the previous context. To me it is both novel and useful.
reminds me of the rabbit hole sessions I used to fall into in https://wiki.c2.com/ (a merge of the two interfaces (chat and context rabbit hole window thing) would be perfect for me)
For me, the snappy, easy to go from one link to another interface reminds me of hours and hours and hours spent browsing https://everything2.com in the earlier years of the Internet.
Perhaps have it not scroll down as it generates the text? Invariably I have to scroll back to the top to start reading. You could have a mini-hud (growing line, with a small rectangle at the top showing the first page of text) which would let you see at a glance how much text is being generated, without interrupting reading. Or not; ChatGPT just keeps on vibrating the phone (iOS app) during text gen, with hovering arrow in the middle-bottom as a shortcut to jump to the end.
ChatGPT just keeps on vibrating the phone (iOS app)
I'd hate that. YouTube turned on vibration (a subtle tap) for videos whenever it reached automatically generated "key concepts" in (some) videos, with no option to disable it[1] so I had to finally disable all vibration on the phone:
Settings -> Accessibility -> Touch -> Vibration (off)
[1] https://old.reddit.com/r/youtube/comments/1bro66c/videos_vib...
You can turn the haptic feedback off in the app, which is what I did. It doesn’t convey useful feedback, just lain annoying.
some feedback:
* can you make it so we can share links of sessions?
* can you describe on the homepage or in a link from the homepage what it does.
Yes, +1 to sharing links! I'll also add:
* Enable the use of personal OpenAI API keys.
* Include system prompts, such as "If the topic is about X, highlight new topics by Y" and "Reply to all as if explaining the topic to a 6-year-old."
* Backlink to the original thread when the same topics are found.
* It would be great if this could be a desktop app with all answers saved locally, creating "my own personal" infinite wiki.
I really, really love this. Even after using it for just a few minutes, I'm sold. Excellent work, will it be open source at some point?
One critique is that I wish there was a feature to manually make a word highlighted. It's pretty good at figuring out where to branch, but for really lexically dense topics, it would be nice to specify it without needing to type in the chat. Perhaps a "hover for definition" feature would suffice, similar to Wikipedia.
I dig this. Reminds me of column mode in MacOS's finder, which is similarly helpful in "rewinding" an exploration of a file system.
Would be interesting to rabbit other rabbit hole resources like Wikipedia or IMDB in this way too.
column mode in MacOS's finder
FWIW these are known as "Miller columns" if anyone wants to research the topic further: https://en.wikipedia.org/wiki/Miller_columns
I really love this. A book was recommended to me that I'm not going to have time to read, but this UI is an amazing way to figure out the main ideas and dive deeper into the interesting ones.
No idea if the things it's telling me are true or not, but that doesn't matter quite as much in this case.
Which book?
This is really cool! I love the rabbit hole stuff you can do when you give GPT more capabilities. I was playing around with this stuff and found I was most often wanting to use it when wanting to learn about something so made Instaclass: https://myinstaclass.com/. It finds videos, images, makes quizzes and gets more relevant web links for you to keep exploring, and structures it like a class (basically a list of bullet points like you mentioned). Try it out and lmk what you think!
awesome work! I've wanted to explore the same idea - glad to see it getting worked on. The chat interface into language models clearly works but it frequently feels like an inefficient way to explore the latent knowledge space of the model. Hypertext (the www) has also been shown to be a great way to explore a massive knowledge space. What this is doing is applying something like a hypertext layer as a way to navigate the model's latent space. Very cool. It could become something of a dynamically generated personalized wikipedia. I'm curious what the prompts look like that you are using to generate subsequent "pages". It could be as simple as "write a wikipedia style summary of <x>" but I think there is a lot of potential in including the context of my current "rabbit hole": "explain <x> in the context of <y> with a learning goal of <z>", etc. Another idea: grounding this kind of hypertext exploration with rag on a specific dataset, e.g., wikipedia or hn.
How does it work?
Are hyperlinks generated as part of original prompt or you do post processing on a response with another LLM?
I've been craving a gpt ui where I could fork conversations that stem from a genesis thought.
Put this on a canvas too so I can zoom out and look at the footprint I left to retrace my steps
This is fantastic, like an encyclopedia that knows what context you are learning about as you skip pages. Nice work!
My minor recommendation is to highlight somehow that the input fields accepts any topic and the suggestions are just random try-it-out topics, it's wasn't immediately clear. Instead of 'write a message' it could say maybe, 'enter a topic to learn about'.
I just get a lightblue screen (Firefox Windows 11)
For coding it would be nice to add task to links. So when a link is clicked you could simply choose to follow it or create a new agent with the link. Each agent will tune the output as one goes down the rabbit hole.
Reminds me of Andy Matuschak's UI as well: https://notes.andymatuschak.org/z9C7piFz8mthkGUUd1W2CR?stack...
crrriispy ui dude. This is like that custom prompt people were sharing for gpt where it would preempt three relevant follow up questions.
what heuristic are you using for making words clickable?
i recommend making the links just hardcode old-school blue and purple. make it obvious you can click these things. "dive on in" https://www.youtube.com/watch?v=DElxVXS7PD0
You may find interesting to look at Google's AI Kitchen and the early versions of Bard (the LaMDA version) because they were specifically optimizing user scenarios where the user wants bullet points (though intuitively, your tool seems already much better than what Google did).
Which LLM is this using because it is providing hallucinated facts that differ from ChatGPT 4o on the latest model?
For some reason I feel like the Wikipedia should use this (or something like this) as a backend to serve a more "dialogish" UI (without replacing the current static UI). The name "delve" is spot-on, but it lacks Wikipedia's intelligence and interconnectedness.
And maybe add some locally stored "Microsoft Recall"-like feature to revisit paths you've made. It would be text-only, so use up almost no space, and be quickly searchable.
Well done it could even work in a terminal.
This is an outstanding idea, I love it!
It would be great if you could support Ollama (or an OpenAI compatible endpoint) for private LLMs.
Really curious about the data structure you're using behind the scenes.
I love the experience. It's almost like a mind map that keeps getting better and better.
rip your api credits
First impression, a fast and neat interface. I went into data indexing rabbit hole as that has been my obsession these days.
Cudos! Particularly impressed with the lack of clutter and the speed.
Source code?
I love how it feels like obsidian
Very nice idea, hope it's not costing you a fortune to run!
Maybe let us put in our own api keys?
How to combine the learning efficiency of this with the structure of a textbook?
Reminds me of the Smalltalk IDE and browser.
Also, WorldSim could use something like this, and perhaps web browsers.
Anyway, I suspect this resonates for anyone that has to do research on the web or in GPT. I often end up with multiple threads on GPT anyway trying to learn about something.
This is cool, it's almost like a wiki you can talk to. I also wanted to make a thread-based UI for LLM chat since I realized that's how I typically interact with them (almost like git branches) but hadn't gotten around to it yet. Neat to see others are interested in branching conversations as well!
Wow, this is really neat! I usually don't comment on Show HN's because I'm rarely impressed by them and I don't want my lack of enthusiasm to be a detterent for people showing their work, but occasionally one like this comes up that is very cool. I also really appreciate that absence of tracking other than Cloudflare Insights (which seems very reasonable to me).
There's an old truism in the business, that the more "suggestions" people give about your idea, the more they like it, and it's absolutely true. Solid work!
Do you have plans to monetize and/or open source it?
I can't believe no one has mentioned Andy Matuschak's work on his notes, "Andy-mode:" https://twitter.com/andy_matuschak/status/156803277302543155...
It's a really clever use of the UI, and I think he'd be happy with it.
Nice work, this is really solid!
I've had something like this on my mind for a while. I really think there are some great use cases for AI around supporting/enhancing human cognition rather than trying to outsource our thinking. In this case of this, being able to rapidly "expand" your working memory with whatever is present in these cards is promising.
I look forward to seeing what you do with this.
I REALLY like how snappy it is. I've always been impressed with how fast Wikipedia managed to stay over the years, but this is even better. Really nice work.
this UI feels right.
Really fun! I realized each delve carries the context of the previous ones. So I got to StarCraft II from the initial example of "Faster language models", but it mostly talked about how SC2 can be used for reinforcement learning. It'd be nice to have a key I can hold down to start a new delve on the topic (bonus points if you can stack multiple delves so you can keep going deeper on the old track as well!)
Another thing that would be interesting is if there was minimal markup for the LLM to indicate "here should be an image of [search term]" or maybe even interactive code blocks etc. But obviously this is scope creep deluxe.
This made me chuckle with delight once I understood what it was doing.
Love this. It would really benefit from a back/forward button though!
I see there are many comments about Wikipedia. I can't find the link, but many years ago (like 7+), there was a concept Wikipedia redesign that proposed the same UX. This was a "marketing" project of a non-Wikipedia related designer. Does anybody remember?
UI is very clean. Left right scroll is awkward without a trackpad, however
Branching conversations are great for a whole bunch of reasons. I posted a demo of a prototype: https://x.com/ajdegol/status/1788689011302682657
And Jake Collins just announced he’s open sourcing an obsidian plugin which has a ton of features: https://x.com/JacobColling/status/1795462258258002255
This is fun! It feels like infinite hyperlinks. It's the kind of wonder I had in exploring Wikipedia for the first time.
This is pretty neat. What I really like is the tiling layout.
I subscribe to phind, which provides a nice search/answer service, which also suggests followup questions, which works fairly well: https://imgur.com/a/WfHSzdk
But if it was in a tiling format, that would be pretty awesome for the flow, especially on mobile.
I used to go down rabbit holes on Wikipedia all the time--could spend hours doing this.
This to me seems like an Infinite Wikipedia! Really cool use-case!
This is pretty great, provides a nice set of breadcrumbs for a deep dive into any rabbit hole.
Once thing that threw me off was when I went to the original panel and clicked a second topic, it cleared out the panes that I had explored off the first topic. I had to discover they weren't really lost by re-clicking. I think it would be better if there was some visual indicator they were still there - perhaps the topic (and sub-topics) get collapsed but are still visible with the heading of the selected topic?
Dark mode please!
i honestly don't get it. what's even different about it than chatGPT?
Click the links.
oh -- it wasn't really obvious they were links. i think i assumed that because i'm used to the chatGPT ui.
They have the underline usually associated with hotlinks
It's a light grey dotted line under a black bold text, it's not impossible to miss.
Funny, I was just thinking yesterday about how back in '90s, ALL links were blue with an underline (or purple if you've visited it).
Not all, but the vast majority yes, because nobody bothered styling links with CSS.
From Tim Berners-Lee webpage:
Rendition of links
Q: I'm a student of visual communications and asked myself why links are blue. I found some answers that might be, for example blue is a color of learning, but I'm not sure what is right. Is there any reason, why links are colored blue ?
A: There is no reason why one should use color, or blue, to signify links: it is just a default. I think the first WWW client (WorldWideWeb I wrote for the NeXT) used just underline to represent link, as it was a spare emphasis form which isn't used much in real documents. Blue came in as browsers went color - I don't remember which was the first to use blue. You can change the defaults in most browsers, and certainly in HTML documents, and of course with CSS style sheets. There are many examples of style sheets which use different colors.
My guess is that blue is the darkest color and so threatens the legibility least. I used green whenever I could in the early WWW design, for nature and because it is supposed to be relaxing. Robert Cailliau made the WWW icon in many colors but chose green as he had always seen W in his head as green.
One of the nicest link renditions was Dave Raggett's "Arena" browser which had a textured parchment background and embossed out the words of the link with a square apparently raised area."
https://www.w3.org/People/Berners-Lee/FAQ.html#your
The dotted underline is usually reserved for indicating alt text or hover content, actually. In this case, I think it's fine to be dotted, since it's not a true hyperlink, but combining that with it being the same text color is just bad from a semantic POV. It's made worse by the fact that the author apparently decided to make visited links blue. (Edit: apparently it's "active" panes, not visited, but semantically similar)
@maxkrieger if you're reading this, please consider making unvisited links blue, to conform to the universal semantics everywhere else on the web, and make visited links either purple, or black if you really want. (edit: or some different color for active panes. Green?)
agreed..although, that's a more appropriate thing to critique to developing a production-ready product than a demo like this.
It's like following the links in wikipedia, but each link is a new chatgpt window to interact with.
Let me see if I can articulate it.
You know how a human conversation can have multiple threads? And ten minutes in, you find the topic has totally changed and you're trying to figure out the original topic? Sometimes you can get back to it, sometimes you can't, right?
Obviously it's not quite the same when you can see prompt history, but the conversation is still pretty linear. This pre-empts that problem by letting you fork thoughts.
also beautiful feature of nested comment threads, like this very orange site :)