Robots Assist Profanity
/@/robots-assist-profanity/Last updated today — Projects Roboletter
Back in high school, my friends were really into cards against humanity. We would sit on the floor in an abandoned stairwell and with a deck teetering in endless depravities. I was never very good at it.
When GPT-2 was released to the public, and I started dabbling in fine-tuning the model on this and that and an idea struck me. What if you fine-tune on every existing card against humanity card and make a deck of entirely generated cards? So I did just that, and soon I had my high school friends were trawling through some 100,000 or so generated cards to find the funniest ones.
The resulting deck was much less obscene and depraved than a typical cards against humanity deck or expansion, it’s humour more benign. It had a certain absurdity to it. I liked it a lot more than the base game. In fact, ever since we have not touched the base set of cards against humanity, instead opting to play with our custom set dubbed “Robots Assist Profanity.”
Funny story, I actually emailed the cards against humanity folks offering to collaborate on an AI-generated expansion. They refused. AND THEN, a couple months later they did do a whole AI-written expansion where they had their writers competing with the AI and getting the community to vote on the funniest cards. The kicker being they were collaborating with another random teenager (roughly the same age as me at the time) to do all this. Frustrating, especially considering my cards were so much funnier than the ones they generated, voted on, and ended up releasing as their expansion.
But maybe it was for the best, in retrospect.