Could you at least mention what the video game was, or why it was such a good implementation? Also, what was "perfect" about the code? "Perfect" is not a word I would ever use to describe code.
This reads like empty hype to me, and there's more than one claim like this in these threads, where AI magically creates an app, but any description of the app itself is always conspicuously missing.
Yes Im exaggerating and its not writing a AAA game from a prompt but I asked it to make a game like Zelda and it figured it out and walked me through all the aspects of it. That's a lot more than I expected. I'm not a games programmer, so I'm probably a lot more impresse than I should be, but I went from not knowing anything about godot to having a framework up to build a 2d rpg-esque game fairly quickly and me learning as it gave me the code. Note, I used the new chatgpt study mode, so that's may be different than just regular prompts. I fully expected just broken code and random AI musings, but instead I got a very solid implementation of a game, albeit a simple one. Or at least as simple as I asked for, I imagine I can keep building out more with its help.
I also have never used godot before, and I was surprised at how well it navigated and taught me the interface as well.
At least the horror stories about "all the code is broken and hallucinations" isn't really true for me and my uses so far. If LLM's will succeed anywhere it will be in the overly logical and predictable worlds of programming languages, but that's just a guess on my part, but thus far whenever I reach out for code from LLM's, its been a fairly positive experience.
Thanks for elaborating, this puts things into perspective, although the complexity of the end product is still unclear to me.
I do still disagree with your assessment, I think the syntactic tokens in programming languages have a kind of impedance mismatch with the tokens that LLMs, and that the formal semantics of programming languages are a bad fit with the fuzzy statistical LLMs. I firmly believe that increased LLM usage will drive software safety and quality down, simply because a) no semblance of semantic reasoning or formal verification has been applied to the code and b) a software developer will have an incomplete understanding of code not written by themself.
But our opinions can co-exist, good luck in your game development journey!
I'm playing with it still and now am adding more scenes and more logic. I think the complexity here is whatever my goals are. I'm not sure what the practical limits here are, or at least they exceed my own ability in games development right now. This is just a toy game, but as I reach into claude and gpt, I can keep going, which is nice. I already have coding experience so I'm not exactly a 'vibe coder' but I think professionally, I dont think people with zero coding experience are getting dev roles, but instead the role will change like my example of the modern mechanic or modern sysadmin above.
As far as QA goes, we then circle back to the tool itself being the cure for the problems the tool brings in, which is typical in technology. The same way agile/'break things' programming's solution to QA was to fire the 'hands on' QA department and then programmatically do QA. Mostly for cost savings, but partly because manual QA couldn't keep up.
I think like all artifacts in capitalism, this is 'good enough,' and as such the market will accept it. The same way my laggy buggy Windows computer would be laughable to some in the past. I know if you gave me this Win11 computer when I was big into low-footprint GUI linux desktop, I would have been very unimpressed, but now I'm used to it. Funny enough, I'm migrating back to kubuntu because Windows has become unfun and bloaty and every windows update feels a bit like gambling. But that's me. I'm not the typical market.
I think your concerns are real and correct factually and ideologically, but in terms of a capitalist market will not really matter in the end, and AI code is probably here to stay because it serves the capital owning class (lower labor costs/faster product = more profit for them). How the working class fares or if the consumer product isn't as good as it was will not matter either unless there's a huge pushback, which thus far hasn't happened (coders arent unionizing, consumers seem to accept bloaty buggy software as the norm). If anything the right-wing drift of STEM workers and the 'break things' ideology of development has primed the market for lower-quality AI products and AI-based workforces.
This reads like empty hype to me, and there's more than one claim like this in these threads, where AI magically creates an app, but any description of the app itself is always conspicuously missing.