When our model saturates a benchmark it's called state of the art reinforcement learning and task generalization, but when theirs does it is called bog standard benchmarkmaxxing etc.
McLuhan wrote how Euclidean geometry was hallucinated from visual space/typography, a subconscious, environmental effect. Jagged, "ugly" math findings would seem to indicate there was something there, bias of visual sense towards the continuous, "beautiful" upper bounds etc
