Complete AI Training
Sign inGet my AI kit

Your job's AI kit

Get your AI kit

Tell us who you are and what you do. We show you your kit right away and email you the link: skills, prompts, AI agents, MCP servers and courses for your job.

500+ jobs ready, and we make a kit for any other job. No payment needed to look.

Share

Video course · 10 chapters · 71 min · certificate

Self-Improving AI Agents: Tools, Code & Constitutional Feedback

How agents improve with tools and feedback: ReAct for combining reasoning and acting, its challenges, RLEF execution feedback for code models, Constitutional AI learning from AI feedback, and experience versus reasoning.

Become a memberAlready a member? Sign in

What you'll learn

  • Explain how ReAct combines reasoning and actions
  • Apply ReAct to grounded knowledge tasks
  • Identify ReAct's challenges with large action spaces
  • Explain reinforcement learning from execution feedback
  • Describe Constitutional AI and AI feedback
  • Compare experience-based and reasoning-based agent training

Chapters

10 chapters · 71:13
  1. 1:47 01Intro Members Building on LLMs LLMs solve language tasks; now we build agents on top of them.
  2. 8:33 02ReAct Members ReAct Humans think before acting; ReAct interleaves thoughts and actions for language models.
  3. 13:00 03ReAct Members ReAct on grounded tasks Using search actions for grounded knowledge tasks.
  4. 12:40 04Limits Members ReAct's challenges Large action spaces and other challenges, discussed with the class.
  5. 5:27 05RLEF Members RLEF: execution feedback for code Reinforcement learning from execution feedback improves code models on competitive programming.
  6. 4:59 06RLEF Members RLEF deep dive The two-tier strategy, loss function and questions on data leakage.
  7. 6:51 07AI feedback Members Constitutional AI Replacing human feedback with AI feedback guided by a constitution of principles.
  8. 12:38 08Results Members Constitutional AI results Results, training sequences and questions about continual learning.
  9. 1:08 09Training Members Experience versus reasoning Collecting experience in real environments versus reasoning about a defined search space.
  10. 4:10 10Wrap-up Members Summary and takeaways Getting LLMs to respond in desired ways is a fundamental building block of agents.