I’ve been having fun using GPT-3 to generate weird versions of internet memes. Even if it’s a meme that didn’t exist when GPT-3’s training data was collected from the internet (which was October 2019 for the version I’m using), most memes will fit within its prompt length and it’s able to follow the format just well enough to be fun, while also bringing in weird stuff because it doesn’t quite get how all this works.
An example meme: people have been inviting their friends to choose their best match among sets of colors and descriptions: “Green – mutual you would probably fight if you met irl”, “Lemon - so cool, they are almost intimidating”. I used a list of these as a prompt for GPT-3 to add to, and penalized it slightly for repetition so I wouldn’t just get repeats of the same common colors. It came up with ideas that are way more interesting in my opinion:
I decided to see if I could influence the list in a particular direction. GPT-3’s job, technically, is to try to fill in the rest of a web page that starts with your prompt text. So what if I listed the original colors, THEN wrote that I’d trained a neural net to produce more but it had become seriously haunted? Turns out it DOES have an effect - here are some of the colors it came up with.
If you want to have a neural net tell you about your personality rather than just giving you options to choose from, may I recommend some AI-generated personality quizzes?
If none of the above are exactly on the mark, I collected a few more interesting colors and descriptions to choose from. Support AI Weirdness and get them as bonus content!
My book on AI, You Look Like a Thing and I Love You: How Artificial Intelligence Works and Why it’s Making the World a Weirder Place, is available wherever books are sold: Amazon - Barnes & Noble - Indiebound - Tattered Cover - Powell’s - Boulder Bookstore