games

The amount of 'generation' happening here - if you take that to mean remixing or creating new geometry or features rather than just a lower-res, non-permanent visual copy of the exact thing it was trained on - is next to nothing. The first corridor, the T junction, the outdoor area, the bridge across it, the corridor into the small indoor arena are all exactly the layout of that section of level in Quake 2. The only difference here seems to be that enemies appear and disappear at random (who wouldn't love to play a game where an enemy shoots you in a corridor and when you try to turn around and fight back it's gone forever?) and when the 'player' gets itself stuck in a dark corner it just teleports them somewhere else and starts again. Oh, and the small indoor arena presumably dead-ends because it's meant to have a staircase around a large pillar in it but the AI doesn't seem to be able to replicate it or deal with verticality in general.

Even scraping the absolute bottom of the barrel of any possible, moderated, benefit-of-the-doubt, ongoing research expectations this is such dogshit to the point of being more of a demonstration of why this tech absolutely will not ever be a viable product. Doubly so when you account for the amount of resources, cost, time, and even people involved in this process compared to what a dozen or so people made 30 years ago and one person could knock up in little to no time comparatively now.