Mar 16, 2024
LLMs are trained to block harmful responses. Old-school images can override those rules.
Gist is, the models have an immune system against you telling them bad stuff. But they (still) don’t have any immunity against you getting them to generate bad stuff themselves. So, this is basically the same shit as “tell me a story like grandma did,” but that was so childish it didn’t seem particularly Star Trek to me; though this is even simpler, it seems specifically at exactly that Trek level. like yesssss.
