When an AI is more creative than Veeky Forums

>When an AI is more creative than Veeky Forums

twitter.com/JanelleCShane/status/977196207863836672

Attached: tg attempts original content.png (584x856, 352K)

Other urls found in this thread:

coinsandscrolls.blogspot.ca/2017/09/osr-1d50-discount-spells.html
emlia.org/pmwiki/pub/web/Tripocalypse.Fauna.html
twitter.com/SFWRedditVideos

Do Date wards protect you during dates or do they deflect the fruit of the same name?

>”Would you let me stat some of these up for 5e, and if so, how would you like to be credited?”
Holy shit what a thirsty tool. All she did was make a simple program to mash two words together. These aren’t copyrighted monsters.

The fruit.

They preserve your wizard powers by stopping you from going on dates

I played around with RNN text generation before. Fun stuff, but you really have to cherrypick the output. The stuff you see posted on twitter is usually pulled out from 1000s of lines of grabage output.

If anyone has suggestions of stuff they want to see generated, I can try training a neural network. But keep in mind that it works best with larger datasets (1000 examples minimum, ideally closer to 10k examples). Also, it’s a huge pain in the ass to scrape data from PDFs, so data formatted in webpages or similar is preferred.

Feed it that creepypasta wiki and tell it to create a new creepypasta.

Seconded

Wrote the spells up here: coinsandscrolls.blogspot.ca/2017/09/osr-1d50-discount-spells.html

Thirded.

That, or just chuck all a few thousand Veeky Forums threads in there (filtering to remove the post #s or it'll just generate numbers). Pick something like the /5eg/, scrape the text, see if it can autogenerate shitposts.

Attached: we_are_not_taking_the_wizard_by_mattrhodesart-dad8dpd.jpg (1224x814, 817K)

It'd be crap.

This depends on what she's using. If its a GAN possibly feasible as you can just let those run instead of over fitting to a result set but there's a catch. Images and 2 strings of worlds are non temporal and the human brain will fill it in. A sentence or worse a story will require that not only is grammar obeyed but that it makes sense reading it from start to finish. There is even a blurb at the end about her attempting this approach with character backstory.

>There was the prince of the sun. He was raised by the arcane arts and accepted him to become a fire work and the pig of the scorpions. He was in the blood of curious by the world to be a part of the church, really with the bartender.

It falls apart because there is no information carried from the start to the end or even past one word.

My guess is that the system in question assigns ID values to each word in the system and trains the NN to produce strings of these IDs and evaluates them on how "real" they seem through human or adversarial agent. For an easier set up Markov chains work at just spitting out related systems in an order that makes "sense".