{"id":164844,"date":"2019-09-27T21:34:48","date_gmt":"2019-09-27T13:34:48","guid":{"rendered":"https:\/\/www.geek.com\/featured\/-1805483\/"},"modified":"2019-09-27T21:34:48","modified_gmt":"2019-09-27T13:34:48","slug":"participating-in-inktober-this-ai-has-a-few-prompt-ideas","status":"publish","type":"post","link":"https:\/\/itteacheritfreelance.hk\/test\/wordpress\/2019\/09\/27\/participating-in-inktober-this-ai-has-a-few-prompt-ideas\/","title":{"rendered":"Participating in Inktober? This AI Has a Few Prompt Ideas"},"content":{"rendered":"
\n\t\t\t\t\t<\/p>\n
(via Skitterphoto\/Pixabay)<\/div>\n<\/figure>\n

It’s the most wonderful time of the year: The first autumn leaves have fallen, which means chunky sweaters, pumpkin spice-flavored everything, and another round of Inktober<\/a>.<\/span><\/p>\n

Every October, artists pick up their pens for the month-long drawing challenge, created in 2009 by illustrator and animator Jake Parker.<\/span><\/p>\n

And every October, Parker posts an official list of daily prompts\u2014a word or phrase to help inspire participants. Artists are not required to use the provided menu, and many make their own.<\/span><\/p>\n

But what would a neural network’s<\/a> Inktober prompts be?<\/span><\/p>\n

Research scientist Janelle Shane trained her pet AI<\/a> on four years’ worth of suggestions, which it gobbled up “like melting an M&M with a flamethrower,” she wrote in a blog post.<\/span><\/p>\n

“My strategy for getting around this was to increase the sampling temperature, which means that I forced the neural net to go not with its best prediction (which would just be something plagiarized from the existing list), but something it thought was a bit less likely.”<\/span><\/p>\n

The results are, well \u2026 mixed.<\/span><\/p>\n

At the starting temperature of 1.0 (“already very high,” according to Shane), the algorithm mostly cycles through the same few words copied from the dataset. Or fills the screen with dots, or with repeated words like “dig.”<\/span><\/p>\n

“Occasionally it generates what looked like tables of D&D stats, or a political article with lots of extra line breaks,” she explained. “Once it generated a sequence of other prompts, as if it had somehow made the connection to the overall concept of prompts.<\/span><\/p>\n

\"\"<\/p>\n

AI-generated Inktober prompts\u2014temperature 1.0 (via Janelle Shane)<\/p>\n<\/div>\n

Turning up the heat in an attempt to nudge the neural net toward original content, Shane found the AI’s phrases<\/a> became longer, though no less incoherent.<\/span><\/p>\n

\"\"<\/p>\n

AI-generated Inktober prompts\u2014temperature 1.2 (via Janelle Shane)<\/p>\n<\/div>\n

At its highest sampling temperature, the machine churned out some of the worst run-on sentences ever:<\/span><\/p>\n

Easily lowered very faint smeared pots anatomically modern proposed braided robe dust fleeting caveless few flee furious blasts competing angrily throws unauthorized age forming<\/span><\/i><\/p>\n

Light dwelling adventurous stubborn monster<\/span><\/i> (OK, so that one almost makes sense\u2026)<\/span><\/p>\n

“It helped when I prompted it with the beginning of a list,” Shane said. “But still, I had to search through long stretches of AI<\/a> garble for lines that weren’t ridiculously long.”<\/span><\/p>\n

\"\"<\/p>\n

AI-generated Inktober prompts\u2014temperature 1.4 (via Janelle Shane)<\/p>\n<\/div>\n

There you have it: Your (very unofficial) 2019 Inktober prompts.<\/span><\/p>\n

Best of luck!<\/span><\/p>\n

More on Geek.com:<\/span><\/i><\/p>\n