In this episode, I sit down with Michelle Pokrass, who leads a research team at OpenAI within post-training focused on improving models for power users: developers using OpenAI models in the API and power users in ChatGPT. We unpack how OpenAI prioritized instruction-following and long context, why evals have a 3-month shelf life, what separates successful AI startups, and how the best teams are fine-tuning to push past the current frontier. If you’ve ever wondered how OpenAI really decides what to build, and how it affects what you should build, this one’s for you. (0:00) Intro (1:03) Deep Dive into GPT-4.1 Development (2:23) User Feedback and Model Evaluation (4:01) Challenges and Improvements in Model Training (5:54) Advancements in AI Coding Capabilities (9:11) Future of AI Models and Fine-Tuning (20:44) Multimodal Capabilities (22:59) Deep Tech Applications and Data Efficiency (24:14) Preference Fine Tuning vs. RFT (26:29) Choosing the Right Model for Your Needs (28:18) Prompting Techniques and Model Improvements (32:10) Future Research and Model Enhancements (39:14) Power Users and Personalization (40:22) Personal Journey and Organizational Growth (43:37) Quickfire With your co-hosts: @jacobeffron - Partner at Redpoint, Former PM Flatiron Health @patrickachase - Partner at Redpoint, Former ML Engineer LinkedIn @ericabrescia - Former COO Github, Founder Bitnami (acq’d by VMWare) @jordan_segall - Partner at Redpoint
