Mitigating Memorization in LLMs: @dair_ai observed this paper presents a modification of the following-token prediction goal named goldfish reduction to help you mitigate the verbatim