How OpenAI Builds for 800 Million Weekly Users: Model Specialization and Fine-Tuning

In this a16z podcast episode, Martin Casado interviews Sherwin Wu, the engineering lead for OpenAI's developer platform, about the evolving landscape of AI models and platforms. They discuss OpenAI's strategy of balancing its API offerings with first-party applications like ChatGPT, addressing potential tensions and competitive overlaps. Wu shares insights on the shift from a single "god model" to specialized models, the importance of fine-tuning APIs, and the surprising stickiness of specific models due to user familiarity and technical optimizations. The conversation also explores the changing views on prompt engineering, the emergence of AI agents, and the complexities of pricing intelligence, as well as OpenAI's approach to open source and the challenges of verticalizing models for different products.