One or both of you may already know about https://aiweirdness.com. It’s a site of GARGANTUAN fun, run by Janelle Shane (Ph.D., I presume, though the site only mentions her Ph.D. studies. It doesn’t specifically say she received the degree and I know too many people who are ABD). Her day job is in optics research (probably playing with lasers, because don’t all those optics researchers play with lasers? Don’t they?), but in her spare time she

train[s] neural networks, a type of machine learning algorithm, to write unintentional humor as they struggle to imitate human datasets.

She’s done quite a lot of this from her places of residence and work, located on the occupied territory of the Arapahoe nation. She’s had computers create heavy metal band names. She’s had them write backstories for Dungeons & Dragons characters. Perhaps most successfully, she’s had them create names for racehorses (most successfully because racehorse names never make any sense, so it’s pretty difficult for a computer to foul up the task). She even attempted to go against type and have a neural network write intentional humor in an experiment where she fed the neural network the text of a large number of jokes and then had the network output its own. The results were NOT fractal. On one level of analysis, the results were clearly predictable. Yet on another level of analysis that clearly did not hold true:

What did the new ants say after a dog?
It was a pirate.

On the other hand, I feel the neural network might have paid me a compliment and/or summed up the inevitable aftermath of every single Jurassic Park movie:

What do you get when you cross a dinosaur?
They get a lawyers.

Important safety tip: Don’t cross the dinosaurs.

“But wait!” you exclaim if you have developed object permanence. “Wasn’t this all supposed to be about zebrafish?”

Why yes, astute reader. This post was and is all about zebrafish. You see, the most recent effort by Dr. Shane was to train networks to take the title of a list and the first few items on a list and then complete the list. For instance, in Olympic events, the 9th place medal is the Sigil of Destruction. Still, that’s better than finishing 11th and getting City Pollen. Shane also experiments with cake ingredients and anniversary gifts, but there is nothing the neural nets love so much as animals. Indeed it gleefully listed its favorite animals, if I am not gratuitously anthropomorphizing. (Fact Check: I am gratuitously anthropomorphizing.) Here you have the favorite animals of neural net GPT-2:

1. Giraffes
2. Maize
3. Polar Bears
4. Pigeons and Giraffes
5. Cats and Warthog
6. Javanese Canines
7. Tiger Teeth
8. Black Swans
9. Alligators
10. Basilisks

and, of course, number

11. Zebrafish