3 days ago · ai

Corrupting LLMs Through Weird Generalizations

Fascinating research: Weird Generalization and Inductive Backdoors: New Ways to Corrupt LLMs. AbstractLLMs are useful because they generalize so well. But can y...