Mitigating Memorization in LLMs: @dair_ai pointed out this paper presents a modification of the subsequent-token prediction goal identified as goldfish loss to help mitigate the verbatim era of memorized instruction data. Perplexity summarization navigates hyperlinks: When inquiring Perplexity to summarize a webpage by way of a link, it navi