In this episode of Unsupervised Learning, Jacob Effron interviews Michelle Pokrass, a key figure behind GPT-4.1 at OpenAI. They discuss the model's focus on real-world utility and developer experience, moving beyond benchmark optimization by gathering feedback directly from users and startups to identify areas for improvement. Michelle shares insights on creating effective evals, the challenges of long-context evaluations, and the importance of instruction following. The conversation explores the current state of AI agents, the advancements in code generation, and the potential for fine-tuning models for specific tasks. Michelle also touches on the future of OpenAI's models, emphasizing a move towards more generalized AI solutions while also discussing the balance between model capabilities and practical implementation for enterprises.
Sign in to continue reading, translating and more.
Continue
