Community
Reddit thread asks whether LLM memory research should focus on forgetting and updating outdated information
A community discussion argues that long-context models fail when information changes, since old and new facts coexist, and asks for benchmarks that test whether models update answers to contradictory evidence.
Every outlet that ran it 1 article
r/MachineLearningCommunityLead
Are we solving LLM memory by remembering more, when the real problem is knowing what to forget?[D]