[TrackerFF]

If we look at the progress made from ChatGPT 3.5 (Nov 2022) and up to today...shoot, I'm really starting to wonder if we'll even be reviewing code in 4 years.

And I'm not saying this as some sort of AI maximalist. If progress keeps up, I seriously doubt software engineering and development will, as we know it today, will be a thing in the next 5-10 years. Maybe humans will be left with designing the UI, but everything else will be abstracted away and AI will be doing all the actual work behind the scenes.

[softwaredoug]

The thing is with a wish machine, you can one shot something very easily.

But then you realize software is the accumulation of 1000s of wishes. And you want this but not that. Many little micro decisions of exactly what you want in every nook and cranny.

In the current paradigm (LLMs) we still have to manage all this. But maybe in the future we have something impossible to imagine.

[weakfish]

You’re assuming that progress will continue at the same rate (it’s plateaud a bit as far as I can tell, I can look for sources tomorrow but am on phone now :))

& I think it’s fairly demonstrated that the industry isn’t profitable for training ever increasingly huge models, which get exponentially more expensive each iteration

[galaxyLogic]

I wonder, LLMs have progressed at great speed, because they have learned to duplicate what humans did and recorded in the documents and source-code that became the training materials for the LLMs. But now that humans don't write any more code, there is no new learning materials for the LLMs. There are no new things for them to learn. Therefore I expect they will plateau, the progress is no longer so fast.

And maybe we'll get something like a Mad Cow Disease, when AI "eats its own brains". To avoid that from happening we will need continued human inputs, and review.

[inigyou]

They are now paying full-time programmers to write new training data.

[galaxyLogic]

In other words humans are teaching the AI.

[zelphirkalt]

Already loathing the idea of not even being able to directly blame humans and their incapability for making shitty websites, that require JS for everything, including typing a text in an input field. If we get there and the LLMs build the same crap, dialed up to 110, because even more suits want first and foremost fancy looks, rather than function, and because that's the majority of training data corpus they learned from ... I might have an existential crisis coming in. The web already sucks hard enough. At least let me blame incapable human beings for this situation.