Search papers, labs, and topics across Lattice.
Tsinghua University, China.
1
0
2
Existing unlearning methods can leak sensitive knowledge through multi-hop reasoning paths, exposing a critical vulnerability in LLMs.
