1 paper
Shashank Sonkar, Richard G. Baraniuk
We introduce Many-Shot Regurgitation (MSR) prompting, a new black-box membership inference attack framework for examining verbatim content reproduction in large language models (LL…