Backdooring Societal Bias into Instruction-Tuned Large Language Models Via Virtual Prompt Injection.

Farsheed Haque, Arpita Ghosh, Depeng Xu, Shuhan Yuan

Browse the full PAKDD paper archive.