Stanford AI Course Map 中文 GitHub ↗
← Back to the course map
Stanford · 2026—27 CS 329A LLMs & Agents ★★★★☆

Core course · LLMs & Agents

Self Improving AI Agents

中文:自我改进的 AI Agent

A research seminar on self-improvement, verifiers, test-time compute, search, tools, multimodal web, and long-horizon planning.

Draft translation: official facts are checked; wording may change after native review.

3 units Spring 2027 Letter or Credit/No Credit ★★★★☆
AgentResearchReasoningSelf-improvement

Course positioning

CS 329A is taught by Anmol Chowdhery and Azalia Mirhoseini and focuses on how agents keep improving through interaction with the environment and themselves. The course is mainly paper discussion and an original research project, with academic and industry speakers; it suits people who have already decided to do agent research.

What it covers

  1. LLM self-improvement: constitutional AI, self-critique, self-play
  2. Verifiers and search: learned verifiers, best-of-n, tree search
  3. Test-time compute: inference-time scaling, chain-of-thought, parallel sampling
  4. Tools and retrieval: tool use, retrieval, multimodal web interaction
  5. Multi-step reasoning and planning: long-horizon decomposition, orchestration frameworks
  6. Evaluation and orchestration: robust eval, failure modes
  7. Applications: coding agents, STEM research assistants, robotics

Prerequisites

CS 224N or CS 229S; fluent Python and LLM APIs.

Assessment & workload

Paper reading/discussion plus an original research project and speaker lectures. Check the current syllabus for details.

Who it's for

Graduate students interested in agent research, alignment, or reasoning/planning; people with some LLM experience.

Study advice

Read ReAct, Reflexion, Tree of Thoughts, and Self-Rewarding LM first. The project needs a clear baseline and evaluation.

Related courses

Official course description

This graduate seminar course covers the latest techniques and applications of AI agents that can continuously improve themselves through interaction with themselves and the environment. The course will start with self-improvement techniques for LLMs, such as constitutional AI, using learned/domain-specific verifiers, scaling test-time compute, and combining search with LLMs. We will then discuss the latest research in augmenting LLMs with tool use and retrieval techniques, and orchestrating AI capabilities with multimodal web interaction. We will next discuss multi-step reasoning and planning problems for agentic workflows, and the challenges in building robust evaluation and orchestration frameworks. Industry applications that will be discussed include coding agents, research assistants in STEM, robotics and more. Students will work on an original research project in this area, discuss the suggested readings in each class, and learn from invited academic and industry speakers. Prerequisites: CS224N or CS229S; Fluency in Python programming and using large language model APIs.