Honestly not that impressive since you can get comparable results with a series of regex rules given that there are limited ways to describe your intent e.g. "create a button of colour <colour> at the <location of button>"
If designers wanted to write texts to create visual designs, they'd be using some form of DSL and learn to code, wouldn't they?
I believe the hype is that people think they can replace the designer by "just telling the computer" what they want. I don't believe that will work, as they already have trouble telling a human what they want, and a computer won't really know what to do with "I want it to kind of feel like it's from that movie with the blue people that Cameron did, you know?"
In my experience, people have a hard time writing their ideas about designs & features down, because they don't know what they want. They want to talk about it abstractly with somebody who has a better understanding of the field so that person can help them develop the idea. I don't think ML will cover that part any time soon.
> people have a hard time writing their ideas about designs & features down, because they don't know what they want
From an academic standpoint, writing is part of the thinking process. If you haven't written it down, you haven't fully thought it through. If it feels difficult, that's probably because your understanding isn't as complete as you thought it was.
From a software development standpoint, implementing something is part of the thinking process. Ever notice how the requirements have a tendency to break as soon as you actually try to implement them? If a spot seems difficult it just means you hadn't really figured it out yet.
> From an academic standpoint, writing is part of the thinking process. If you haven't written it down, you haven't fully thought it through.
I 100% agree. I noticed a giant shift in tasks when I made one client write tickets instead of making phone calls. Writing it down forces you to think it through.
And I agree about software development as well, yes. Though I think it's even rare to have somebody describe all the features they want unless it's an experienced software developer who basically writes a textual representation of the application.
But for most PMs (that I've worked with at least), they have vague ideas about what they want, and bringing them into focus is a back and forth with developers and designers. I don't see them getting anywhere with an NLP automaton, but maybe with an Eliza-style system: "Give me a big yellow button saying 'Sign up'" - "Why do you want a big yellow button saying 'Sign up'?" - "You're right, that's too on the nose... give me a link saying 'Sign up'"...
Honestly not that impressive since you can get comparable results with a series of regex rules given that there are limited ways to describe your intent e.g. "create a button of colour <colour> at the <location of button>"