MemoryBank: Enhancing Large Language Models with Long-Term Memory

Summary

Proposes a long-term memory mechanism for language models that recalls relevant memories, updates them continually and builds an understanding of the user’s personality; inspired by the Ebbinghaus forgetting curve, it lets memories fade or strengthen according to the time elapsed and their relative significance.

CommentaryA direct translation of human forgetting into AI memory — and a prompt to ask whether fading should change what exists, or only what is brought to mind.