Trapdoor Prompts and the Hidden Behaviors of Language Models
-
A trapdoor prompt is an input designed to trigger a specific output from a
language model, without using any of the words in that output. It’s not a
gues...
9 hours ago
0 comments:
Post a Comment