Janelle Shane, a brilliant writer and researcher whose work explains our increasingly machine-intermediated world in ways we mere humans can understand, is back with another neural-net-generated collection of candy hearts, this time using GPT-3. Previous iterations have gifted us with STANK LOVE, SWEAT POO, LOVING HORN, and, naturally, BALLS.

One thing I didn’t realize (still awaiting my API token!) is what we know as GPT-3 is actually a collection of different neural net models of varying capabilities. Shane tested with several variants, to excellent effect.