I'm sure this line of logic is very comforting, but frankly, this comfort disappears quickly when you actually
use ChatGPT. What you find is that you can interact with it in a quite natural way, and it is able to synthesize and iterate at a level that feels easily on par with a moderately skilled human software engineer. I know it's uncomfortable, but it doesn't even matter if the machine is "non-intelligent." Nobody gives a damn. What matters is what you can do with it, and every iteration of GPT the goal posts keep moving further, but this time it's really difficult to deny: you really, really can describe a program at a high level and ChatGPT can implement it. You can point out an error and it can fix it. Hell, you can feed it compiler errors.
Is it literally as good as a human software engineer? No, but it's also better too. I doubt ChatGPT could debug as effectively as a veteran software engineer (... In fairness, most humans can't either.) It can debug pretty decently, but there's still work there. That said, the breadth of knowledge encoded in a language model is stunning. I'm pretty sure you can't just regurgitate an implementation of the discrete cosine transform in Rust without at least pulling up Wikipedia, but ChatGPT can, because well. It doesn't have to pull it up.
I still don't think ChatGPT is ready to replace human programmers. It may be a long time before we have general enough intelligence to replace knowledge work meaningfully with AI. However, if you think it's not happening ever, because machines are not "intelligent" based on some set of goal posts, I've got bad news: that's not part of the job listing.
It's easy to laugh at MSN publishing articles written by GPT; that's just stupid. However, at some level you have to admit that the input to ChatGPT is almost as high level as directives from project managers, and the output is almost low level enough to simply input directly into source control. That leaves very little to the imagination for how this could quickly spiral out of control.
Is it literally as good as a human software engineer? No, but it's also better too. I doubt ChatGPT could debug as effectively as a veteran software engineer (... In fairness, most humans can't either.) It can debug pretty decently, but there's still work there. That said, the breadth of knowledge encoded in a language model is stunning. I'm pretty sure you can't just regurgitate an implementation of the discrete cosine transform in Rust without at least pulling up Wikipedia, but ChatGPT can, because well. It doesn't have to pull it up.
I still don't think ChatGPT is ready to replace human programmers. It may be a long time before we have general enough intelligence to replace knowledge work meaningfully with AI. However, if you think it's not happening ever, because machines are not "intelligent" based on some set of goal posts, I've got bad news: that's not part of the job listing.
It's easy to laugh at MSN publishing articles written by GPT; that's just stupid. However, at some level you have to admit that the input to ChatGPT is almost as high level as directives from project managers, and the output is almost low level enough to simply input directly into source control. That leaves very little to the imagination for how this could quickly spiral out of control.