It seems likely that the entirety of Zork (world state and the possible actions to transform it) is already learned by the model. Which means that there is a grue in there, too. Not good. I’m starting to re-think the doomer argument...
Perhaps this is a stupid/contentious idea (partly because it somewhat kills the "spirit" of the original games), but there's a little part of me that would be interested in seeing the scene building parts of Zork piped into an image generation service to visualize the landscape that the game describes.
(the grue would obviously just a picture full of black, though some creepy eyes would be a nice touch)
