Poking (fun) at the content filters of Dall-E 3 and GPT-4

submited by
Style Pass
2023-11-19 12:30:03

Content warning: cartoon insinuations of blood, a joke about the crucifixion of Jesus, freudian jokes, copyrighted characters, cartoonish background images of destroyed cities.

In my circles a lot of people are currently busy finding applications for generative AI, by founding startups or just polishing their resumes as deep learning experts and prompt engineers. I think the hope is to extract one more modicum of value before the whole thing blows up in the face of every knowledge worker in the labour market.

Not me, I’m not enough of a hustler, so instead I’m interested in the weird bits, and questions like: How do companies make their model(s) output only socially acceptable content? How good are the safeguards? Surely they trained the dog to not pee on the carpet, but they can’t completely prevent the occasional mishap, right?

So the latter point might put a big “duh” on this article. You could question the usefulness of another “look here at the unhinged output I made an AI produce”. So why is this still interesting?

Leave a Comment