Is the claim that these models can 1 shot a Simpsons episode remake with different camera angle and similar dialog from a prompt like "produce Simpsons episode S01E04"? Or are we falling into the "the user doesn't notice that they told the model the answer, and the model in fact did not memorize the thing" trap?