FRESH Hacker News
Home
Weird Generalization and Inductive Backdoors: New Ways to Corrupt LLMs
5 points by joegibbs