I think they're aware of how absurd their claims are, but they think that the most important thing is the "potential" for this technology to generalize and eventually come to achieve what they're promising now. The issue is they'd essentially need AGI to take it from this to a real game engine, and that's obviously not happening.