The video says "No coding skills? No worries. Just write your code with ChatGPT."
I also wonder if they'd even know what to ask.
He's been leaning heavily on the repl.it Copilot thingy, so it's more autocomplete than code blocks. He uses ChatGPT for working out error messages.
Person A: I just used AI to make my own website, you're gonna be out of a job soon.
Person B: Oh? Let me see it.
Person A: Here it is C:\Users\Bob\Desktop\index.html
If you can already code, then GPT can ramp you up on a new project faster, or accelerate and existing project
I was raised with (80s/90s); never break the public api unless you have to and make a migration path if you do; now it is ‘let’s definitely break the public api, with any minor version change and just waste everyone’s life fixing it, we don’t care’. And gpt doesn’t know this. It usually does make up code that’s close to correct and relatively easy to fix though, but you need to know how to read the docs and not panic that actually everything is spitting errors.
If you ask it for C, Java or vanilla js it fares better but there you suffer from the memory window; now with 16k tokens, we are seeing massive improvements. The api deprecation (read; we just throw away or change without warning) issue remains and is a huge issue; I cannot even see how they would properly fix that in LLMs. They are going to have outdated versions in their search space; how do they know to ignore them? And next to that there are the hallucinations.
To me, it's like a junior developer I have to coach constantly.
The result is definitely not something you'd want to maintain and probably has built in inefficiencies and edge cases I don't know about but it did it's job, which was to do the thing I wanted it to do so I can go on with my work.
Sometimes I think “you can tell this part of my code was generated by ChatGPT because it’s actually good, unlike the rest of it”
Of course it’s likely to make subtle mistakes that are hard to debug if you ask it to do anything complicated because it can’t test its own code (well, I’ve heard something about a code interpreter in the Plus version but I don’t know how well that works)
The problem was that there were breaking API changes across versions of the library, but it was just using probabilities of method names being correct, and it didn't ask me or tell me which version of the library it was targeting, so it was just a mismash of incompatible code.
But the code LOOKED good and the AI was confident it was a good answer. Overall, it felt like wasted time sifting through confidently-wrong answers.
it works ok as a starting point of " I have no freaking clue how to approach this" to "ahh I can see how that will work" and then you go write your own code with the lead.