NVIDIA's VP of AI Explains Why They Give Away Their Best Models | Kari Briski × Kim Isenberg

In this interview, Kari explains the hybrid Mamba-Transformer architecture, why NVIDIA trained in FP4 precision from day one, how speculative decoding delivers up to 3-5x speed gains, and why NVIDIA — unlike other labs — releases full training recipes, datasets, and environments to the open-source community. Plus: the newly announced Nemotron Coalition with Mistral AI, the Ultra base model reveal, and why NVIDIA's business model makes radical openness possible.