Hacker News Reader

Comment by littlecranky67 | Hacker News Reader

Because it was not trained on screenshots or real rendered computer UIs, but text. That is also why in my experience, LLM suck at describing click paths, and are less helpful on UI development, as they never really "see" the result of the code as in rendered HTML outputs.

View on HN

The terminal is also just the easier way to instruct someone to do things. "Just run this" is easier than a step by step guide through UIs which often change.

I go into the shop, I walk up to the counter, and I say "Can I have a 1/2" drive T50 Torx bit please", and the person behind the counter says "Yes of course" and we go over to the small expensive tools cabinet and get one out.

I don't go into the shop and wander about until I find something that looks like it, then stand there pointing things going "THAT!" until someone figures out what I mean.

And now I have a T50 Torx bit that I can stick on a ratchet with a long extension and get the passenger seat out of the Range Rover so I can retrieve my daughter's favourite necklace from where it's gotten entangled with the wiring to the gearbox and suspension ECUs in a place where I can see it with a dentist's mirror but can't actually get a grabber onto to fish it out, worse luck.

So that's my afternoon sorted then. Because we're not just hacking on computers round here.

On the other hand, if you went and browsed the visual interface, you might discover you could purchase a 1/2” drive to 1/4” hex adapter, thereby opening up the possibility of using the entire set of impact driver bits you already own.

That doesn't solve the problem I have, because I already have a 1/4" ratchet and I don't have a 1/4" T50 bit.

Furthermore, a T50 bit with 1/4" drive would just snap instantly. If the bit didn't break, you'd twist the end off the extension bar.

I have a specific problem, which I already know how to solve, which has a specific solution, for which I need a specific component.

Nonsense. They make T50 bits for regular impact drivers and they don’t immediately self-destruct.

Anyway the point was about the discoverability factor of user interfaces.

So why would a 1/2" to 1/4" adaptor help, then?

Consider that to crack the seat bolts free initially, I need to use a metre long breaker bar to get them moving, and I need to torque them down to something ridiculous on refitting.

When Windows 95 was introduced as a fully graphical operating system every manual coming with Microsoft software instructed you to open the "Run"-dialog and type your drive letter followed by "setup.exe" to install the software.

Like the past several decades of Linux problems, you find some stack overflow answer saying just run this command. Terminal is eternal, UI changes

Getting people in the habit of running random commands they don’t understand in the terminal seems dangerous.

The UI is also less stable then most cli tools.

The enterprise tools I am currently working with often have outdated screenshots in their own documentation.

Alternative take: Because no designers are getting paid to move "rm" to "fileops rm" or otherwise between releases.

Getting them to take screenshots with playwright/puppeteer and look at them as part of their development iteration cycle works well.

For local inference, sure, but we simply lack the computing power to train them on all the images and html content that is available in the internet and books. That will happen sometime in the future, though.

Ah right, sorry, you were making a much more interesting point than my reply! I read "UI development" and jumped to the conclusion that the point was just about inference-time modify-test cycles. Yes, agreed, if they trained on images, or even better (?) on (code, image) or (code-delta, image-delta) pairs, they would surely be better at UI development.

Yep and also their 'click paths' (<- love that by the way) are trained on READMEs which are often out of date.

Reply on news.ycombinator.com