Some awards show controversy had House House disclose possibly the lightest possible LLM usage you can imagine. Anyway it’s a Bad Game now.
Amongst all the glut of genuinely vibe coded slop, inventive AI reverse engineering and the plain ol’ realities of LLM living in modern software development, I get a bit frustrated seeing it all grouped into one essentialized category.
The common understanding of what AI even is, is at an all-time low, imo. People see ‘AI code’ next to a game title and imagine someone prompting claude for 20 minutes, and these sorts of stories and over-the-top reactions don’t help.
The basics of what happened with this story, is that a particularly bad Unity bug plagued the team and the unity support wasn’t being a help. LLM fixed the bug. I’m shocked to see that even this usage is considered unacceptable to some people.
I’m shocked to see that this usage is considered acceptable to some people.
What have they really done wrong here? The solution could have been delivered by Unity support eventually and it would likely be the same fix. They could have accidentally been presented snippets of code simply googling aspects of the bug, because google search automatically runs your search through their AI. They also could have not engaged with the LLM and shipped the game with the bug intact, or made other concessions, and the end result would have been lesser.
I am asking this in good faith despite how opinionated I already seem about it. I just don’t ‘get’ the thought process and why we would expect a rational human being to behave differently in this situation. Solidarity and all that, yeah - but it’s not like they’re enthusiastic AI developers. They claim AI was used in no other part of development and what they produced was one of the most human and expressive video games out there.
So from where I’m standing, such resolute opposition just feels dogmatic, without really considering the current realities of software development. I’m half-way between ‘this is an appropriate use of the technology, and surrounding market dynamics are the problem’ and ‘good boy developers get a little LLM as a treat’
I basically agree with you, but
is such a weird and needlessly weasely way to talk (and you do this in the OP too). You aren’t giving any useful new information, just saying “that’s just how things are now” in a more pretentious and knowing tone. If you are trying to say anything rather than just engage in sophistry with this phrase, you are failing.
It really does feel like “AI realism” similar to “Capitalist Realism” I hear this argument in pretty much every article on AI, “Yeah, it’s actively worsening society and harming people and killing the planet, but that’s just how things are now so there’s nothing we can do to change it.”
The current reality of software development for me is that I had to fully and forcefully rip out all AI generated code in anything that I actually use last year and rebuild it from scratch.
I get the allure of “have the stochastic parrot fix it for me”, but you can just add easily spend some time fixing it yourself. It you’re able to prompt one of these things in a way that actually solves the problem, you’ve already solved the problem.
Since you know about this and I do not, could you explain how something could happen like the OOP? It seems like a few coders tried to fix it, both the devs and whatever poor intern was doing Unity support or however that works.
I should be clear, I don’t mean the particulars of this case, but some people apparently not being able to figure it out and an LLM doing it. Would you call this exclusively explicable as laziness or time-saving or something of the sort on the part of those people?
I’m not completely sure. There was probably a pressure to fix it, they probably tried and fell back on the LLM because they didn’t have the time or couldn’t be bothered.
They didn’t give much info on exactly what the code was. Either way, you can almost always solve these issues with dev time unless they’re actually unsolvable (which this one isn’t, since we know they found a patch).
I just know that in the past, whenever I did that a monkeys paw curled. The accumulation of tech debt becomes exponential and you lose your ability to navigate and understand the codebase.
When you write everything, or write on a team, you start understanding the ways that people solve things. You can navigate the code just by remembering landmarks. LLMs take that away, and you quickly enter a situation where you get instantly lost and unable to fix things yourself.