zlacker

I feel like I am missing something or holding it wrong, I would personally love if we had a tool that i could describe problems at a high level and out comes a high quality fully functional app. Most software is shit and if we are honest with ourselves there is a huge amount of inessential complexity in this field built up over the years. I would gladly never spend weeks building something someone else already built in a slightly different way because it doesn’t meet requirements, I would gladly not end up in rabbit holes wrestling with some dependency compatibility issue when I am just trying to create value for the business. If the tools get better the software gets better and the compexity we can manage gets larger. That said while these tools are incredibly impressive, having messed with this for a few days to try to even do basic stuff, what am I missing here? It is a nice starting point and can be a productivity boost but the code produced is often wrong and it feels a long way away from automating my day to day work.

>>mrbomb+(OP)
> I would personally love if we had a tool that i could describe problems at a high level and out comes a high quality fully functional app.

I'm sure your employer would love that more than you. That's the issue here.

> That said while these tools are incredibly impressive, having messed with this for a few days to try to even do basic stuff, what am I missing here? It is a nice starting point and can be a productivity boost but the code produced is often wrong and it feels a long way away from automating my day to day work.

This is the first irritation of such a tool and it's already very competent. I'm not even sure I'm better at writing code than GPT, the only thing I can do that it can't is compile and test the code I produce. If you asked me to create a React app from a two sentence prompt and didn't allow me to search the internet, compile or test it I'm sure I'd probably make more mistakes than GPT to be honest.

>>kypro+w1
If I had the tool that did that, I would be the employer!

>>alt227+f2
Exactly code has always been a means to an end not the end itself. Further our industry has been more than happy to automate inefficiency away from other fields, feels pretty hypocritical to want it to stop for ours.

>>kypro+w1
Have you actually tried to get an app working using gpt? A lot of shared stuff is heavily curated. It is no doubt an extremely impressive tool but I think we always underestimate the last 10% in AI products. We had impressive self driving demos over a decade ago, we are all still driving and L5 still seems a ways away.

>>alt227+f2
I mean, then so would everyone else, and we just fall back to a capital problem to advertise your creations.

>>pixl97+9V
If everyone is able to make their own app, then there is no need to advertise their apps, because everyone will just be using their own.

The real battle there would be protocols; how everyone's custom apps communicate. Here, we can fall back to existing protocols such as email, ActivityPub, Matrix, etc.

>>mrbomb+(OP)
I think a lot of people are coming at this from an all or nothing approach. Does the current solution do everything we keep expecting ML to do? No, does it do way more than before and get us x percent closer? Yes.

Think of the 80/20 model, if it gets you 80% there (don't take that literally) then that's huge in it of itself. This tool is getting us closer to the example you mention and that in of itself is really cool.