I normally wouldn't inform you of a public slapfight between a developer and X users, but unfortunately, given the absolute fury-storm surrounding generative AI technology, and the fact this whole debacle's an illustrative point as to why all conversations surrounding it are tainted by ragebait, my hands are tied.
Tim Soret—who got in deep water after espousing support for GamerGate years ago, then backtracked, then promised a comeback of his game The Last Night in 2021 (you will observe it is now 2025)—has repeatedly and emphatically stated on Twitter that the first ever known novel invention of AI occurred: A somewhat-pretty generation that looks like a dressed-up version of Daggerfall.
Interesting to see the praise for this AI-invented visual style. Everybody wants it, and indie devs are rushing to prototype it & make it real. https://t.co/rASfMaZpxMJuly 21, 2025
"If no human has ever produced this particular blend of techniques before, then yes it has just been invented," Soret goes on to insist in a chain of increasingly agitated comments, adding that he did the same thing with The Last Night back in 2017. I want to point out, in the gentlest way possible, that Octopath Traveller was also announced in the same year—and games take years to develop. Sometimes ideas happen in parallel, and it's nobody's fault.
Anyway, Soret's general misunderstandings about how AI actually functions is a decent cautionary tale to not allow the jangling keys of generated work fool you. Especially when they involve getting into arguments on X—anyway.
I'm not stupid enough to sit here and tell you that this generation doesn't look cool, but generative AI doesn't think, for starters. Don't take my word for it, listen to someone far smarter from MIT, senior paper author Ashesh Rambachan (in the full story I just linked).
"We see these models do impressive things and think they must have understood something about the world. I hope we can convince people that this is a question to think very carefully about, and we don’t have to rely on our own intuitions to answer it."
Secondly, as we've seen in plenty of videogame tech demos, AI can hold on to a 3D environment for about two to three seconds before it becomes a disturbing, non-Euclidian slurry. In particular, when you turn into a dark corner, AI has a hard time remembering the room you were in and will often "put" you somewhere completely different the moment you turn back around. This is why the videos Soret is praising are a few seconds long.
In other words, even if you want to argue that this artstyle wasn't created in a far more rudimentary form back in the '90s, the AI hasn't actually invented anything because there's no way to make it work right now. In the same way that I have not invented a new airplane by drawing a pretty good picture of one. What I have invented is a picture of an airplane.
In this case, human artists had already made pixel art and gothic fantasy. Was the AI the first thing to go 'man, wouldn't it be cool if we could do this'? Well, I'm almost certainly sure not. At the bare minimum, that was the person who promoted it to cook these videos up—and that's giving them a lot of credit.
So, are these videos valuable as a baseline concept for actual development? Well, kinda, maybe? By far the most frustrating thing about generative AI (aside from the catastrophic climate concerns, or reckless layoffs in favour of pushing tech that isn't fit for purpose yet) is the fact that the actually-useful applications get drowned out in a sea of justifiable outrage.
Machine learning isn't the devil, people are. Larian's using AI for testing human ideas before then moving on to actually making stuff by hand, or heck, back in 2018, Into the Spiderverse used a deep learning model to help artists draw and its comic book lines.
The videos linked by Soret are, at best, an interesting proof-of-concept that I'd like to see actual artists, and game developers, produce something similar to some day. Complete with actual intent, vistas framed by people who understand environment design, and I dunno, maybe an actual game mechanic or two. Something more than just vibes.
But no, either AI has to be the next coming of techno-Christ, or mercilessly used to try and rinse all human creation out of videogames. Soret cannot simply say "this looks cool, let's talk about how we could actually make it"—he has to praise the divine gifts of the Omnissiah and argue that programs MIT says can't think, in fact, do. I'm very tired.
