Meta FAIR introduces a method for AI agents to improve their own memoryMachine translation
Meta FAIR 提出让 AI 智能体自行改进记忆的方法
Self-Improving Memory lets an agent discover prompts for managing file-based memory, then trains those skills into the model. Across three continual learning tasks, the optimized prompts averaged 56.0, versus 44.6 for the best human-written prompts; the trained model scored 43.1 using generic prompts. The tests covered new instances of the same tasks, and code has not been released.
This source permits summary display only.
Read at the original source