Closing the Loop Between AI Training and Inference with Lin Qiao - #742

This podcast episode from Practical AI features Lin Qiao, CEO of Fireworks AI, discussing the importance of aligning AI training and inference systems. The core argument revolves around the need for a seamless production pipeline, moving away from treating models as commodities and towards viewing them as core product assets. The episode highlights post-training methods like reinforcement fine-tuning (RFT) for continuous improvement using proprietary data. A key focus is on "3D optimization"—balancing cost, latency, and quality—guided by clear evaluation criteria. The vision is a closed-loop system for automated model improvement, leveraging both open and closed-source model capabilities.