I build ChatGPT-2D with the idea to enable people to branch AI conversations interactively, pose contextual questions based on AI response and visualize your entire dialogue on a 2 dimensional map interface. What do you think?
This is a really good idea, I wish I had thought of it. Thinking of chat as a graph instead of a list is very weird and very cool.
I think this could be especially useful in a research context where someone is trying to evaluate the performance of a model. Sometimes you just want to compare how a model responds with and without certain context, and doing so with the classic UI is cumbersome.
To be clear; what you say you're interested in seeing is having this graph structure (a singly rooted n-ary tree) expressed as a, and I quote “a vector space 𝔤 together with an alternating bilinear map 𝔤 × 𝔤 → 𝔤 that satisfies the Jacobi identity”
Yea I want to see the 2d non-spatial map have spacialized properties within the self-contained g-space.
That is right. Is saying it in big words somehow more confusing for you?
Basically Jacobi identity would be about things like disambiguations on Wikipedia so again… are you just intimidated by words you don’t know or something? That makes me sad for you.
the jacobi identity is [x,[y,z]] - [[x,y],z] = [y,[x,z]] ie. it says the lie (pronounced "lee") bracket is almost associative but you have to add a [y,[x,z]] when you rearrange brackets. it has absolutely nothing to do with disambiguations.
It's a tree laid out more-or-less intuitively (like a mind map) on a two dimensional plane which someone is calling a map (you can quibble with the term "map" if you like but it's not a mind-bendingly incomprehensible term) in this context. It's not a power set, there would appear to be multiple singly rooted n-ary trees. You start with an initial query (the root node) and the generated ChatGPT responses (they are called cards here) branch out like a tree rather than being a linear back-and-forth convo – structured like, as I said, a mind map https://en.wikipedia.org/wiki/Mind_map
It could also just be used to understand basic concepts with detail. The subcategories and different "priority" you can make with this mindmap layout is strong!
You have probably one of the best landing pages I have seen in a long time for a new product. There's no BS, in 10 seconds I can understand what the idea is about. Well done!
It was two-hour work (with the help of AI, of course). I was inspired by Midjourney (credit to them). But, yeah, as a nerd, I do not know how to bullshit. So I thought it's easier (for me) just to put the demo video and a tagline. Didn't even bother to write a description.
I've been working on something very similar in the past three weeks. Just out of curiosity, what have you used for the layout?
In my project I used reactflow for all things graph and dagre for the layouting (together with some custom hooks to make it work with variable size nodes).
I'm not sure if I will try to monetise it or if I will simply open-source it. Probably the latter because when building time is over and marketing time starts, I get very bored and lose interest soon.
I'm really enjoying the Superus-101 playground for newbies! I've been playing around with it for a while now, and I find it really helpful for brainstorming. However, I did notice that the waiting time for a response can be a bit long. Do you think it would be possible to implement streaming and concurrent modes for generating new nodes in the graph?
Hey! Glad you're enjoying the Superus-101 playground ! The waiting time (ragard a single thread) is actually dependent on OpenAI, with avg wait around 8-15 sec , not much I can do there. But, concurrent threads? Cool idea, definitely worth exploring! Thanks!
Very cool! Could definitely see myself using this. Out of curiosity, what's the visualization library you're using to generate the boxes/lines/zoom in-out canvas? Or is that all native to your app? Building something unrelated and looking for something like this
I did this manually myself to explore some topics I was curious about: I'd copy&paste responses into a Google doc, and then add sub and subsubsections with answers to sub-questions.
Whimsical, a diagram tool, has a new feature that works a little like this but you have to do a bit more work, I've been enjoying it. I'll have to give this a go, thanks op!
I like this. I've done something similar by asking chatgpt to map out some ideas for me. It knows how to output into the `dot` format which makes the result easy to post-process.
Haha, The fact people are taking time to complain is a good sign of engagement.
Looks like fixed spacing along x and letting the cards float along y which is very effective.
I have sometimes wondered why there isn't more convergence for a standard set of primitives/libraries/template engines for drawing graphs on the web.
graphviz .dot files would seem to be a natural DSL and then there are packages like dagre/mermaid but italways seems like for applications you need to end up rolling your own.
However Pina Coladas are easy to make and a delicious summer treat, with or without alcohol. To make them extra special, cut open a coconut and use a stick blender to make them fresh in the shell!
They require very few ingredients, and none of the ingredients are special to just making cocktails, making Pina Coladas an easily accessible drink to create.
They are also, at their heart, dirt cheap. Canned coconut cream, heat and mix in some sugar to make a coconut cream syrup if so desired, or if you don't mind it less sweet skip this step. Add in some canned pineapple, and some rum, and you have a basic version down.
For reference, here is what happens when I ask ChatGPT for a pina colada recipe in a conversational tone
Here's what you'll need:
1 1/2 oz of your favorite rum
2 oz of pineapple juice
2 oz of coconut cream
1 cup of crushed ice
A slice of pineapple (if you have it)
A cherry (if you want to be fancy)
And here's what to do:
1. Put all of the ingredients (rum, pineapple juice, coconut cream, and crushed ice) into a blender.
2. Blend everything together until it's nice and smooth.
3. Pour the mixture into a glass.
4. If you want to make it look extra special, put a slice of pineapple on the rim of the glass and add a cherry on top.
5. Sip and enjoy!
If you like your drinks on the sweeter side, you can add some sugar or simple syrup to the blender before you blend everything together. And if you don't have coconut cream, you can use coconut milk instead, but the drink won't be as creamy.
----
In general, it completely fails at being conversational at all about it!
not what i was saying whatsoever. saying they write the code, and then they collect monthly service fees for something they already did. The SAS model is way over done, and imo this is an example of reiterates that.
I built a successful one, and sure - it gave me a decent income for 10 years after the initial 1 year of working on it. But altogether it was around €0.5M over it's lifetime, and it took me 5-10 years of unsuccessful projects to finally do one that succeeds.
It is a common experience as well. That's where the saying comes from that "it takes 10 years for an overnight success".
Also, the difference is in payout. When you work for someone, you get paid monthly. When you work on your own thing, you get literal 10 years of people making fun of you and saying "why don't you find a real job", and then - if you succeed - you get the same people complaining that you drink pina coladas.
I've been to the bay area, and it's the most work-intense environment I've ever seen. There is no easy money there.