A distraction from interesting worlds that makes players play GPS-find-the-objective(s) instead of exploring just for the fun of it.

My other theory is this is also why they're having trouble with Beyond Good & Evil 2. It's just SO FAR off the vibe of the original.

As someone who used to work there and also as a big fan of some of their older games including that one: I think the problem is Ubisoft is all-in on gritty over-detailed realism these days and Sands of Time is a highly stylized game and those two things cannot be married together.

Sounds like he's conflating several things there: Practical vs CG, animation & rendering. There's nothing inherent to Unreal that makes it look worse than Maya. You can do bad lighting in Maya too. A lot of CG looks bad because its overdone or wasn't given enough time. Not because of software used.

Lovely.

Also they have the sense of wonder and lack of experience to appreciate a game fully on its own merits without caring about whether it's 'formulaic' or just like this other game or whatever. Any game can be somebody's first game, or first favourite game.

True, but a lot of people are conflating the two and lumping ethical small scale machine learning in with the big AI plagiarism energy wasting garbage and that's unfortunate.

Yeah, but (and I could be wrong here) it sounded to me like OP was claiming this and things like it don't and cannot exist.

Yeah, there's no such thing as a 'clean' LLM, OP is right about that. All large AI models are contaminated with stolen work. Only ethical AI is the kind trained from scratch exclusively on selected material used with permission, which by definition cannot be a large model.

Can't use that for concepts but you can do other stuff. Like here's a 2017 example of machine learning AI that literally only does animation transitions on any terrain from a specific set of motion. It doesn't have a base model, it's just a bunch of mocap and the AI figures out the best transitions.

It's possible Vincke is confusing the two though I doubt it as his background is in programming, but there's really a lot of misinformation going around about this and it sounds to me like they would only use custom small-scale machine learning but people don't understand what that means.

Machine learning AI has existed for decades and has been used for various purposes and trained on small specific data sets. Only these giant internet size models are new and THOSE cannot be done ethically without stolen material and so they shouldn't be done at all.

These people are talking about two different things. You cannot make a general purpose LLM or giant image model without tons of scraped data, but you can absolutely make smaller machine learning AIs tailored to specific purposes as people have been doing for ages.

Really glad I had my real camera with me and only took a couple shots with the phone cause wow, it just ruins the image.

I mean look at this shit (photo taken with phone, zoomed in):

See also auto AI sharpening (on by default if you can even turn it off at all), that makes any slightly blurry photo look like it was AI-generated.

"Slopsumers" maybe, but doesn't have the same ring to it.