Weird Generalization and Inductive Backdoors: New Ways to Corrupt LLMs

7 pointsposted 2 months ago
by joegibbs
(arxiv.org)

No comments yet