1 兴趣方向
看了这篇综述《PERSONAL LLM AGENTS: INSIGHTS AND SURVEY ABOUT THE CAPABILITY, EFFICIENCY AND SECURITY》，智能个人助手设计的基础能力中，记忆能力更令人感兴趣。
里面提及了，通过日志、原始数据的管理和处理、记忆增强、agent自我革命（Agent Self-evolution.）等手段实现llm基础能力之一——“记忆（memorizing）”的构建。
Agent Self-evolution包括Learning Skills(利用 LLM 中嵌入的先验知识，并利用执行反馈允许语言模型调整其选择。)和Finetuning LLM”
“Agent Self-evolution. To better accommodate users, Personal LLM Agents may also need to dynamically update themselves based on the memory data. We refer to this as “self-evolution”. The foundational functionality of intelligent agents is predominantly reliant on LLM. Therefore, the key to the self-evolution of intelligent agents lies in how to leverage LLM for the discovery and exploration of new skills, as well as in the continuous update of the LLM itself.”
2 总体认知：
“We first discuss the capabilities required by Personal LLM Agents to support diverse features. Excluding the general capabilities of normal LLM agents, we focus on three fundamental capabilities for personal assistants, including task execution, context sensing, and memorization.
Memorization (§4.3) is to record the user data, enabling the agent to recall past events, summarize knowledge and self-evolve. While context sensing and memorization are abilities associated with querying information from users, task execution refers to the ability of providing services to users.
Memorizing denotes the capability to record, manage and utilize historical data in Personal LLM Agents. This capability enables the agents to keep track of the user, learn from past experiences, extract useful knowledge, and apply this acquired knowledge to further enhance the service quality. The related work is mainly aimed to answer two questions, including how to obtain the memory and how to utilize the memory. ”——摘
摘要：
智能个人助理 （IPA）:
我们设想 Personal LLM Agents 将成为即将到来的时代最终用户的主要软件范式。
讨论了有关 Personal LLM Agent 的几个重要问题，包括它们的架构、能力、效率和安全性
