Earlier I tried an experiment where I tried to get the GPT-3 neural nets,
trained on a huge collection of internet text, to generate new Disney princesses
[https://aiweirdness.com/post/649266848537427968/the-next-disney-princesses].
From just two example wiki entries, I got incorrect reimaginings of existing
princesses (Snow White apparently has
![New Disney villains](/content/images/size/w795/image/fetch/w_1200-c_limit-f_jpg-q_auto:good-fl_progressive:steep/https-3A-2F-2Fbucketeer-e05bbc84-baa3-437e-9518-adb32be77984.s3.amazonaws.com-2Fpublic-2Fimages-2F64e14f28-66f7-447b-b92d-338ac523b8c6_1494x624.png)