Overview
Explore a 16-minute technical video lecture delving into LLM agents and their reasoning capabilities through the lens of core research developments. Learn about Chain of Thought prompting techniques, Program-aided Language Models (PAL), and the ReAct (Reason + Act) framework used in Langchain and CrewAI agents. Understand how these methodologies enable LLMs to interact with external tools while improving their reasoning abilities. Progress through detailed explanations of tools, plugins, functions, and APIs, examining practical implementations of ReAct in JSON/XML formats and fine-tuned models. Discover OpenAI's Function Calling and modified ReAct approaches like Browser agents and CodeAct. Conclude with a comprehensive overview of limitations and important cybersecurity considerations, supported by academic references and research papers for further exploration.
Syllabus
- Table of Contents
- Chain of Thought Prompting
- PAL Program-aided Language Models
- ReAct Reason + Act
- Tools, Plugins, Functions, APIs
- ReAct in Practice JSON/XML formats, fine-tuned models
- Function Calling OpenAI
- Modified ReAct Browser agents, CodeAct
- Summary
- Limitations & Cyber Security Considerations
Taught by
Donato Capitella