Hacker News new | past | comments | ask | show | jobs | submit login

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.




It is a nice UI and invites you to investigate more...

But the problem, as far as I can tell, is that it's inviting someone to explore what's bad about LLMs (or what LLMs are bad at).

IE, LLMs are useful for doing things an individual could do but doesn't really want to. I have one friend who uses ChatGPT for boiler plate nondiscrimination policies and another who uses it for random villain descriptions and it's famous for boiler plate code.

But using LLMs for discovering new specific things (this app's seeming purpose) seems like a recipe for disaster. For example, I started looking at counterfeit bolts and ended up with the thing hallucinating an instance of "sword net" (real) that in 2018 targeted counterfeit fasteners (no refs on Google, Brave or DuckDuckBing) with the slogan "Secure the Foundation, Eliminate the Fake" (no refs similarly).

Edit: obviously, the system is confusing counterfeits generally with counterfeit fasterners (a more specialized issue, having less to do with intellectual property as such). But if drill distinctions like this are inevitable and this is what makes LLMs actually not useful for this sort of exploration.


The way I think about it is LLMs are good at DOING for you, and poor at THINKING for you.


> I can't put my finger on it, but it seems more important than a mere "gimmick."

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.


What I really need for ChatGPT is to ask questions on a side panel and not push out the message exchange.


counterpoint: the forks dont retain any of the context that led you to them, nor does returning to an earlier branching point retain the discussion that occurred down a separate "rabbit hole". therefore it is in some ways decidedly less human that the linear approach in use


They do? When going into “weight” coming from “aerodynamics -> flight” it only talks about weight in the context of flight and plane design. I would actually like an option to “snap out” of the current topic.


So perhaps an even better interface would be a dynamically generated spider diagram?


also beautiful feature of nested comment threads, like this very orange site :)


i honestly don't get it. what's even different about it than chatGPT?


It's like following the links in wikipedia, but each link is a new chatgpt window to interact with.


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


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 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




Consider applying for YC's Spring batch! Applications are open till Feb 11.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: