Mitigating Memorization in LLMs: @dair_ai mentioned this paper offers a modification of the next-token prediction aim known as goldfish reduction to help you mitigate the verbatim era