Weird Generalization and Inductive Backdoors: New Ways to Corrupt LLMs

Image from Unsplash

Weird Generalization and Inductive Backdoors: New Ways to Corrupt LLMsJan Betley, Jorio Cocola, Dylan Feng, James Chua, Andy Arditi, Anna Sztyber-Betley, Owain EvansAbstract: LLMs are useful because they generalize so well. But can you have too much...

This snippet was provided by a publicly available feed. Read the full article at eyepool.tumblr.com