In this video, Sam Witteveen discusses OpenAI’s recent update that introduces structured outputs in JSON format, enabling guaranteed structured responses from their API. This update enhances the existing functionality of function calling and allows developers to receive JSON responses consistently. Witteveen explains the significance of structured outputs for building LLM agents, particularly in tasks that involve data extraction from articles. He demonstrates how to use the Pydantic library in Python to define data models for extracting information such as people’s names, product names, and organizations from text. The tutorial includes a practical demonstration where Witteveen showcases how to scrape articles and extract structured information, emphasizing the importance of constraining outputs to improve accuracy. He also discusses the cost reductions associated with the new model and highlights some limitations and restrictions regarding JSON schemas. The video serves as a comprehensive guide for developers looking to leverage structured outputs for more effective LLM applications.