Stanford Webinar - Agentic AI: A Progression of Language Model Usage

For more information about Stanford's Artificial Intelligence programs visit: https://stanford.io/ai In this webinar, you will gain an introduction to the concept of agentic language models (LMs) and their usage. You will learn about common limitations of LMs and agentic LM usage patterns, such as reflection, planning, tool usage, and iterative LM usage. This online session will cover: Overview of LMs LM Usage and limitations Retrieval Augmented Generation (RAG) Tool usage Agentic LMs Agentic design patterns About the speaker: Insop Song Insop is a Principal Machine Learning Researcher at GitHub Next. Previously he worked at Microsoft, where he focused on leveraging machine learning and large language models to boost engineering productivity. His projects included fine-tuning open-source large language models with internal code and text, developing document assistance tools, and applying AI to various engineering tasks. He is currently a course developer as well as a course facilitator for Stanford Online’s professional AI program. Chapters: 00:00 - Introduction 00:10 - Overview of the Talk 01:50 - Training Language Models 02:30 - Modeling Objectives 04:00 - Examples of Training Data Formatting 05:40 - Applications of Language Models 06:50 - Using API for Language Models 09:00 - Best Practices for Prompt Preparation 11:10 - Importance of Clear Instructions 13:40 - Reflection and Improvement Techniques 16:30 - Tool Usage and Function Calling 20:30 - Definition of Agentic Language Models 21:50 - Reasoning and Action in Agentic Models 24:00 - Example of a Customer Support AI Agent 29:20 - Summary of Applications 36:00 - Key Design Patterns in Agentic Models 44:00 - Summary of Agentic Language Model Usage 47:40 - Audience Q&A 50:00 - Addressing Ethical Considerations 54:50 - Getting Started with Language Models 57:00 - Resources for Staying Updated 58:20 - Closing Remarks