课程定位
CS 224G 是一门项目课,把 LLM 当成新的“CPU”,探索围绕它构建应用的方式。课程以两周一个 sprint 的节奏推进,配工业界 guest lecture,适合想快速做出可演示产品的人。
主要内容
- LLM 能力与局限:幻觉、上下文、成本
- Prompting:结构化提示、few-shot、角色设定
- 工具与检索:function calling、RAG、外部 API
- 应用架构:前端、后端、流式输出、状态管理
- 快速迭代:sprint、用户反馈、指标
- 项目:做一个 LLM-native App
先修
编程能力;基本 ML/LLM 认知;不要求研究背景。
考核与工作量
两周一个 sprint 的作业 + 最终项目 + 客座讲座。具体以当季官网为准。
适合谁
想快速做 LLM 产品/创业的人;产品经理、设计师、全栈工程师。
学习建议
和 CS 329Z 互补:224G 更偏产品节奏,329Z 更偏系统与 eval。先想清楚用户问题,再选模型。
相关课程
官方课程描述(英文原文)
With ChatGPT, neural networks have had their Lisp moment. Conversation has become code and the model is the CPU for this ultimate programming language. A new universe of App development has opened up, and there are no guides for it, yet. This is a project course designed to explore the space of Apps built around LLMs, starting by playing with them, learning their limitations, and then applying a set of techniques to program them efficiently and effectively. Assignments are due on a two week "sprint" cadence to mimic a startup style environment. Guest lectures by area experts provide industry perspective.