Peng Yu
Titles as the sources gave them. In 1 entry, with 0 checked quotations.
In the Base
Every entry they appear in.
DACE: Diversity-Driven Adversarial Co-Evolution for Robust LLM Safety Alignment
According to its abstract, a preprint by Peng Yu, Xiaoyu Wen and seven co-authors introduces DACE, a training framework in which an attacker and a defender model evolve together. It aims to stop the attacker narrowing to a few tricks and the defender forgetting earlier attacks.