斯坦福 CS329 · AI 学习资料包 · 双主线学习地图

整体架构双路径独立发展，于「世界模型 / 物理 AI」汇合

两条独立主线：① 语言/智能体自改进；② 视觉/多模态/机器人感知。各自演进，最终在通用智能处整合。

路径一 · LLM / 智能体自改进

自我改进与推理

CS229 / CS224N基础ML + NLP 地基

CS324大模型基石Percy Liang · 教科书级

CS329TAgentic 应用RAG + Agent 架构 + 评测 + 安全

CS329H偏好 / 对齐RLHF · DPO（Sanmi Koyejo）

CS329A自我改进旗舰测试时计算 · 搜索 · 记忆 · 评测 ⭐

已建资料包：34 篇论文 + 9 节视频｜点击下方各节课直达

路径二 · 视觉 / 多模态 / 机器人

感知与具身

CS229 / CS230DL 基础Andrew Ng

CS231N视觉核心Fei-Fei Li · CNN→ViT→CLIP→扩散→世界建模

CS224N语言半边VLM 的文本侧

CS223A / 225A / 327A机器人感知Khatib 控制流派 · 具身

多模态 Agent · VLM · 具身 AI感知模块 + 规划/控制

独立路径 · 详见路径二_多模态视觉与机器人_学习路径.md

🧩 融合点 · 世界模型 / 物理 AI / 通用智能

两条路径在此汇合

世界模型统一感知(视觉) + 预测(环境) + 行动(规划/控制) + 语言(LLM) → 物理 AI / 通用智能体。

DreamerV3 (Nature 2025)Genie 3 (DeepMind)NVIDIA CosmosWorld Labs (Fei-Fei Li)AMI Labs (LeCun)

路径一提供「自我改进 / 推理 / 记忆」↔ 路径二提供「物理世界理解 / 感知 / 规划」→ 合并为 通用智能体

核心资料 · CS329A 按课进度论文本地可点开 · 视频在线观看

📋 课表/论文清单 🎬 视频清单 🧭 推荐学习序列 🛰️ 路径二·多模态/机器人

L01

Course Overview

Mon Sep 22

Part 1Course Overview

本次无必读论文（客座讲座 / 项目周）

L02

Test-time Compute Scaling

Fri Sep 26

Part 2Test-Time Compute Scaling

📄 Large Language Monkeys: Scaling Inference Compute with Repeated Sampling (Brown et al. 2024)📄 Archon: An Architecture Search Framework for Inference-Time Techniques (Saad-Falcon et al. 2024)📄 Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters (Snell et al. 2024)📄 How Do Large Language Monkeys Get Their Power (Laws)?

L03

Robust Verification

Mon Sep 29

Part 3Robust Verification

📄 Shrinking the Generation-Verification Gap with Weak Verifiers 📄 Training Verifiers to Solve Math Word Problems (Cobbe et al. 2021)📄 Let's Verify step by step (Lightman et al. 2023)📄 Math-Shepherd: Verify and Reinforce LLMs Step-by-step without Human Annotations (Wang et al. 2023)

L04

Learning from feedback with tools/code

Fri Oct 3

Part 4Learning from Feedback with Tools/Code

📄 ReAct: Synergizing Reasoning and Acting in Language Models (Yao et al. 2022)📄 RLEF: Grounding Code LLMs in Execution Feedback with Reinforcement Learning 📄 Constitutional AI: Harmlessness from AI Feedback

L05

Multi-step Reasoning/Planning

Mon Oct 6

Part 5Planning and Multi-Step Reasoning

📄 SWiRL: Synthetic Data Generation & Multi-Step RL for Reasoning & Tool Use 📄 Language Agent Tree Search Unifies Reasoning Acting and Planning in Language Models (Zhou et al. 2023)📄 SPRINT: Enabling Interleaved Planning and Parallelized Execution in Reasoning Models 📄 ADaPT: As-Needed Decomposition and Planning with Language Models (Prasad et al. 2024)📄 Wider or Deeper? Scaling LLM Inference-Time Compute with Adaptive Branching Tree Search

L06

Train Time Scaling/Scaling RL

Fri Oct 10

Part 6Train Time Scaling / Scaling RL

📄 STaR: Bootstrapping Reasoning With Reasoning (Zelikman et al. 2022)📄 DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models 📄 DAPO: An Open-Source LLM Reinforcement Learning System at Scale

L07

Open-Ended Evolution of Self-Improving Agents

Mon Oct 13

Part 7Self-Improvement and Deep Research Agents

📄 Automated design of agentic systems 📄 The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery (Lu et al. 2024)🔗 AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms

L08

Self improvement with Search & Deep Research Agents

Fri Oct 17

无公开视频

📄 Competition-Level Code Generation with AlphaCode 🔗 AlphaCode 2 Technical Report 📄 Search-o1: Agentic Search-Enhanced Large Reasoning Models

L09

Guest Lecture Melvin Johnson (Google DeepMind)

Mon Oct 20

无公开视频

本次无必读论文（客座讲座 / 项目周）

L10

Mid term presentations

Fri Oct 24

无公开视频

本次无必读论文（客座讲座 / 项目周）

L11

Mid term presentations

Mon Oct 27

无公开视频

本次无必读论文（客座讲座 / 项目周）

L12

Mid term presentations

Fri Oct 31

无公开视频

本次无必读论文（客座讲座 / 项目周）

L13

Agentic Frameworks for Software Engineering

Mon Nov 3

无公开视频

📄 CodeMonkeys: Scaling Test-Time Compute for Software Engineering 📄 KernelBench: Can LLMs Write Efficient GPU Kernels?📄 Improving Parallel Program Performance with LLM Optimizers via Agent-System Interfaces

L14

Augmenting Agents with Memory Guest Lecturer: Junchen Jiang (LMCache, UChicago)

Fri Nov 7

无公开视频

📄 Cartridges: Lightweight and general-purpose long context representations via self-study 📄 MemGPT: Towards LLMs as Operating Systems (Packer et al, 2023)📄 CacheBlend: Fast Large Language Model Serving for RAG with Cached Knowledge Fusion

L15

Guest Lecture Denny Zhou, Google DeepMind

Mon Nov 10

无公开视频

本次无必读论文（客座讲座 / 项目周）

L16

Guest Lecture Thang Luong, Google DeepMind

Fri Nov 14

无公开视频

本次无必读论文（客座讲座 / 项目周）

L17

Agentic Evaluations & Long-Horizon Tasks

Mon Nov 17

Part 8Agentic Evaluations and Long Horizon Tasks

📄 Measuring AI Ability to Complete Long Tasks 📄 GDPVal: Evaluating AI Model Performance on Real-World Economically Valuable Tasks 📄 DeepScholar-Bench: A Live Benchmark and Automated Evaluation for Generative Research Synthesis

L18

Guest Lecture Misha Laskin (Reflection AI)

Fri Nov 21

无公开视频

本次无必读论文（客座讲座 / 项目周）

L19

Guest Lecture Danny Driess (Physical Intelligence)

Mon Dec 1

无公开视频

本次无必读论文（客座讲座 / 项目周）

L20

Future Research Areas

Fri Dec 5

Part 9Future Research Areas

本次无必读论文（客座讲座 / 项目周）

整体架构 双路径独立发展，于「世界模型 / 物理 AI」汇合

自我改进与推理

感知与具身

两条路径在此汇合

核心资料 · CS329A 按课进度 论文本地可点开 · 视频在线观看

Course Overview

Test-time Compute Scaling

Robust Verification

Learning from feedback with tools/code

Multi-step Reasoning/Planning

Train Time Scaling/Scaling RL

Open-Ended Evolution of Self-Improving Agents

Self improvement with Search & Deep Research Agents

Guest Lecture Melvin Johnson (Google DeepMind)

Mid term presentations

Mid term presentations

Mid term presentations

Agentic Frameworks for Software Engineering

Augmenting Agents with Memory Guest Lecturer: Junchen Jiang (LMCache, UChicago)

Guest Lecture Denny Zhou, Google DeepMind

Guest Lecture Thang Luong, Google DeepMind

Agentic Evaluations & Long-Horizon Tasks

Guest Lecture Misha Laskin (Reflection AI)

Guest Lecture Danny Driess (Physical Intelligence)

Future Research Areas

整体架构双路径独立发展，于「世界模型 / 物理 AI」汇合

核心资料 · CS329A 按课进度论文本地可点开 · 视频在线观看