自我学习代理
自我学习代理是一种能够根据工作过程中发生的事情进行改进的代理:你做出的修正、任务的结果、你接受或拒绝的编辑。它会改变自己的行为,而不仅仅是改变上下文。
自我学习代理是一种根据工作过程中发生的事情来改变工作方式的代理,而不仅仅依赖于你当下告诉它的内容。
这种学习是狭窄且具体的,这正是它有用的原因。它不会重新训练模型。它会将学到的内容记录为记忆,并在下次应用:
- 一次修正。 你说“不,这里用箭头函数”。这变成了一条规则,并且从那时起一直生效。参见 规则与事实。
- 一个任务结果。 在实际工作中什么有效,什么需要重做,会影响它处理下一个任务的方式。参见 代理如何在工作中学。
- 你保留的内容。 你接受和拒绝的编辑是表明你需求的信号,这就是为什么结果会改善,而你无需编写规格说明。
有两个边界值得了解。它只在记忆开启的表面上学习,而不会在 隐身聊天 中学习,后者在设计上是只读的。而且它学到的东西归你所有,你可以检查并删除:参见 更改或忘记记忆。
Related articles
- Agent memoryAgent memory is what an agent stores between sessions and brings back when it is relevant: standing rules, facts, studied material, and short summaries of past conversations.
- AI agent with a memoryAn AI agent with a memory is one that keeps what it learns about you and your work, attaches it to your account rather than to one app, and applies it wherever you use the agent next.
- How an agent learns as it worksDropstone is a self-learning agent. It records what it learns from corrections, from finished work, and from sources you ask it to study, and it uses that the next time without being told. Here is the loop.
- Rules and factsDropstone keeps two kinds of memory. Standing rules apply to every task without being searched for. Facts are recalled only when they are relevant. Here is why the difference matters and how it decides.
Ctrl+I