Search papers, labs, and topics across Lattice.
Dongguk University-Seoul
2
0
3
Unlearning harmful content in diffusion models can now be achieved without sacrificing the generation of benign co-occurring concepts, thanks to the innovative ReCARE framework.
SlotGCG reveals that the position of adversarial tokens can dramatically increase jailbreak attack success, outperforming existing methods by a significant margin.
