I spent quite a bit of time using gpt-2 to try and make some new Dungeons & Dragons spells. I noticed that around 1 in a thousand spells it would generate would kind of make sense if you interpret it as being trying to be funny. It seems like AI is better at making funny things (unintentional so).

Here are two examples of some spells it generated.

Arcanum creates a massive pool of magical energy similar to that created by the Chariot of Crustaceans. The energy travels down the caster's legs and retains a magical aura similar to that of the ancient people of Cyprus. The spell creates one of the following: a dense fog, raging waters, strong winds, continuous thunderstorms, or a strong gust of wind. The material components are sand and powdered animal hoof.

Holding out his palm, the caster evokes a magical array of sounds in the area where he stands. The following sequence of events occurs, based on the alphabetical order of the alphabet. 01 Stand Long Time 02 Last Words 03 Reach Down From The Ground 04 Make A Friend At A Distance 05 Five Minutes Friends 06 Friends For 10 minutes 07 A Word For Theeunday 08 Move Large Object 09 No Talking 10 Naga Tribe 11 Soundtrack To A Favourable Weather 12 How To Be Seen By A Smart Elemental 13 Back To The Main Team

I decided to collect the best ones and put it in this free pdf if anyone wants to read some more.

https://www.drivethrurpg.com/product/324417/Clockwork-Demon-...

> It seems like AI is better at making funny things (unintentional so).

Partly this is because we take AI-generated sentences slight differently to ones by humans maybe. It feels less funny once we found out someone behind intentionlly making up the joke, while AI never be "intentional".

Yeah, in theory it's possible there was some AI text generation used at some point, but it's clearly writers aping a style, a lot of the funny 'errors' are not something a text-prediction AI is likely to make (especially with how punny most of them are). The rest of the series is pretty funny, though I think the horror movie is the best one.

It's Keaton Patti, a comedian whose shtick is writing nonsense and pretending that it was written by an AI.

So it's all just a marketing trick? There is no AI behind this, just a comedian?

Like most AI companies, it is just people.

Don't mind the man behind the curtain.

Yep, I mean amazon literally calls its service "mechanical turk" x)

From the last time I looked into it (I may be recalling this incorrectly): the script is AI generated text that the writer sifted through, edited, and stitched together to form an amusing, vaguely coherent narrative.

So it's perhaps closer to "inspired by AI" than "written by AI".

most of the comments in this comments section are written by Keaton Patti to stimulate discussion. Including this one

It’s Twitter.

I googled the name and couldn’t find anything on that. How do you know?

Janelle Shane, who does this sort of thing for real on the excellent blog https://aiweirdness.com gave a good breakdown on Twitter: https://twitter.com/janellecshane/status/1007061610005794817

It's mentioned in the article.

"Netflix worked with Keaton Patti, who's done this bot thing before. This time round the bot watched over 400,000 hours of horror movies before penning its great masterpiece, 'Mr. Puzzles Wants You to Be Less Alive'."

Right, but that’s not the assertion. The assertion is that it’s not really a bot, which I couldn’t find anything about.

They remind me of Half Life: Full Life Consequences[1], which wasn't written by a bot but pretty much has all the qualities of a 'written by bot style video

https://www.youtube.com/watch?v=OHxyZaZlaOs

IIRC, it is the guys who made it took a a poorly written Half Life fan fiction (possibly written by a 9 year old), brought to life by a gary's mod machinima

Sigh, yeah, the CNet journalist^W writer is either incompetent or lying... if he thinks "It's just a silly article", does he and his editor think CNet is The Onion, without letting us know? Or can I consider CNet a publisher of fake news?

I think you can safely consider cnet an unreliable news source.

Maybe it’s an ad. Paid for

Could be the bot wrote it, but like many movies etc, someone wrote the screenplay, based upon the story.

Lots of change from story to screenplay...

My "That 70's Show", sci-fi reunion story would suffer too greatly in this process, so I am scared to send it to Netflix.

Well, they never said the bots they are using also use the latest AI technology. It just says "bots".

These could be purposely flawed bots, with AI only good enough to create rubbish, hilarious, fantastic rubbish.