Hacker News Reader

Comment by valine | Hacker News Reader

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.

View on HN

[weird and cool] that's the best compliment for any hacker! Thx mate, you just made my day

This has been done a few times in the past few months, here’s a nice one: https://creativity.ucsd.edu/ai

I’d be interested to see this expressed as a Lie (pronounced “Lee”) Algebra personally.

https://en.m.wikipedia.org/wiki/Lie_algebra

Why?

can you elaborate?

About what in particular?

It seems you’re making a https://en.m.wikipedia.org/wiki/Hierarchy_(mathematics) which isn’t dimensional in any nature. It’s like saying that hacker news is a “2d forum”…

Anyway ordering it more like https://en.m.wikipedia.org/wiki/Power_set

Just seems natural to me. Information typically isn’t hierarchal outside of artificial structures.

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

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”

Riiight

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.

hope that helps lol

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!

Think it's great

Reply on news.ycombinator.com