As a software developer, I think GPT-3 & friends and beyond is the real next-gen step-change in software development, not SwiftUI. In 10, 20 years, I could see a future where that’s how all programming is done, with ‘low level' languages like SwiftUI(!) acting as merely the assembly language beneath high-level prompt-based compilers. Programming languages just an autogenerated build artifact. Voice-based holodeck programming come to life
(definitely not a technology to sleep on, or downplay)
@stroughtonsmith and you will do it all from your flying car or jet pack.
@ToddLa I've been doing it from my desk, but whatever works for you!
@stroughtonsmith Already had it write very useful starting points for several little scripts/projects for me. It’s not perfect, but with a few back&forth iterations, it gets a fair bit of the way there, ready for me to take over. Super impressive.
@stroughtonsmith Regardless of the syntax specifics, me and my coworkers are already using it everyday, all day. I feel like I just got an upgrade as a human.
@stroughtonsmith and if you accidentally include the wrong things in your training dataset, you end up with Moriarty trying to escape its sandbox and take over the ship! 😅

@stroughtonsmith I can't wait to see the future where a language model sits at an overarching level in the IDE

Not only suggesting code, but new file names, assets, colour palettes. Designing and running instrumentation, tests, localization and documentation, and then suggesting deployment scripts

@stroughtonsmith It’s gonna be the evolution of no code and WYSIWYG to… WYSIWYG (what you SAY is what you get)
@stroughtonsmith 11 years ago I started out iOS development on a no code platform. Couldn’t believe the progress I was making in a day - until I wasn’t. These abstractions are written by people and they make mistakes - and when I hit enough bugs I started with the transition to ObjC. They limit what you can do also, have the same feeling with SwiftUI, every app just looks the same.

@stroughtonsmith I imagine they are great for creation and duplication, but hard for modification though:

"See that green button that you see when you do X, Y, Z (and a dozen options and actions)? Maybe it red instead"

"Not that one, the button to its top right"

"No, no, a bit more red"

So the low level bits still has to stay reasonable high levels for human, unless we get to the next step in its evolution?

@stroughtonsmith I’m trying to wrap my head around what the future looks like though. What happens when the results has to feedback into the training set because people don’t do enough non-gpt programming anymore?
@stroughtonsmith in my understanding these language models need input and as long as there isn’t great input you won’t get great output either. So whatever new languages and technologies might come along, you’d first need humans that use those and produce some output for a next level language model to ingest and be useful in this new area too. But maybe I’m just not getting some magical “all languages are basically the same on a conceptual level” kind of theory underneath it all. 🤔
@stroughtonsmith I partly agree, but I don’t think we’ll reach that point until the ML models are trained to produce code that meets specifications, rather than just producing code that is a plausible continuation of what’s been typed so far. The experiments with Bing and ChatGPT have shown that the current technology produces confident liars. The more output you ask for, the farther it drifts from the original intent.
@stroughtonsmith Meanwhile back in 1981.... The future is not what it used to be.
THE LAST ONE

@stroughtonsmith SwiftUI is incredible. GPT-3 is cool but it’s only as useful as what you feed it ie the code it steals from developers.
@stroughtonsmith Remind me SwiftUI, it is all cool and fun for small demo, it get increasingly painful when you need to do something different, and eventually you will have to port it back to UIKit.