No transcript available for this episode.
This episode examines the March 2026 model release cycle, where GPT five point four functionally tied Gemini three point one Pro Preview at the top of the Intelligence Index while seven of nine releases shipped as open weight. We analyze mixture of experts architectures achieving active parameter ratios as low as two point five percent, enabling three hundred ninety seven billion parameter models to run inference at ten billion parameter cost profiles. Coverage includes Grok four point twenty Beta’s twenty two percent hallucination rate, Qwen three point five’s eight variant product line spanning mobile to data center deployment, and MolmoWeb’s pixel coordinate based visual agent architecture. The episode closes with analysis of the White House National Policy Framework for Artificial Intelligence, its federal preemption proposals, and congressional response across partisan lines.