Stanford AI 课程地图 EN GitHub ↗
← 返回课程地图
Stanford · 2026—27 CS 224V LLM 与 Agent ★★★★☆

核心课程 · LLM 与 Agent

智能体 AI

Agentic AI

把 LLM 变成能查、能推理、能执行长任务的 agent:RAG、工具、知识库、形式化方法、长程规划。

3–4 学分 2026 秋 Letter or Credit/No Credit ★★★★☆
AgentRAGLLM项目

课程定位

CS 224V 是一门“项目课 + 硬核讲座”的 agentic AI 课程,由 Yuhao He 等人讲授。它不只教怎么调 API,而是系统讨论 agent 的可靠性问题:如何减少幻觉、如何混合结构化/非结构化知识、如何用形式化方法保证决策、如何优化长程任务。

主要内容

  1. RAG 与任务形式化:检索增强、formal task descriptions、减少幻觉
  2. 混合知识推理:数据库、知识库、非结构化文本联合推理
  3. AI 驱动知识 curation:科学发现中的知识整理与发现
  4. 形式化方法:提高决策 agent 的准确性与可解释性
  5. 长程 agent:自动优化长程任务的准确率和效率
  6. quarter-long project:方法研究或领域 agent 构建

课程大纲 / 课件

#主题 / 讲者
01Introduction: LLMs hallucinate — turning them into dependable agents
02Knowledge Curation: STORM / Co-STORM
03Research project ideas & AOS architecture
04Accountable AI: SMT, SatIR, VERDICT
05Student project ideas
06Constraint Satisfaction Based Retrievers
07Constraint Satisfaction Based Retrievers + Project Proposals
08How to RAG without hallucination: WikiChat
09Project Proposals
10ReAct Agents for Querying Knowledge Bases and Graphs
11Agentic Data-Driven Research: DataSTORM
12Querying Hybrid (Structured and Unstructured) Data: SUQL
13Long-Context Long-Horizon Deep Researcher: SLIDERS
14Minimizing hallucination in task-oriented agents: Genie Worksheet
15Coding Agents: anatomy of Claude Code / Codex
16VLMs for Transcription of historical manuscripts
17Training LLMs: instruction following and training data

Fall 2026 schedule;slides 在 Canvas,公开页只放 schedule 和主题。

先修

LINGUIST 180/280、CS 124、CS 224N、CS 224S、CS 224U 之一。

考核与工作量

讲座 + 一个占全年很大比重的 quarter-long project(研究型或应用型)。具体以当季官网为准。

适合谁

想做 agent 研究、AI 科学家、领域自动化的人;已经会 LLM 基础、想深入可靠性和形式化保证的人。

学习建议

先熟悉 ReAct、Toolformer、RAG 和 function calling。项目尽量选一个可量化的任务,否则很难做 eval。

相关课程

官方课程描述(英文原文)

AI agents powered by Large Language Models (LLMs) are already transforming how we work, learn, and solve problems - and we are only at the beginning. As these systems grow more capable, they hold extraordinary promise for accelerating scientific discovery and democratizing access to high-quality medical, legal, and educational services worldwide. This is a project course coupled with rigorous lectures on the principles, methodologies, and cutting-edge research underlying agentic AI. Students undertake a substantial quarter-long project in either foundational methodology research or building novel agents in a domain of their choice. Topics covered include: (1) minimizing hallucination in question-answering and task-oriented agents using Retrieval-Augmented Generation (RAG) and formal task descriptions; (2) hybrid knowledge reasoning over databases, knowledge bases, and unstructured text; (3) AI-driven knowledge curation and discovery for scientific research; (4) improving the accuracy and interpretability of decision-making agents through formal methods; and (5) automated techniques for improving the accuracy and efficiency of long-horizon agents. Prerequisites: one of LINGUIST 180/280, CS 124, CS 224N, CS 224S, or CS 224U.