Episodes from No Priors: AI, Machine Learning, Tech, & Startups about Open-Weight AI Models.

The New Apex Predator: CrowdStrike CEO George Kurtz on Adversaries Armed With AI Agents
Oct 22, 2026 · 45:28
CrowdStrike CEO George Kurtz tells hosts Sarah Guo and Elad Gil that AI agents have flattened the adversary pyramid: the 'agent state' gives hacktivists nation-state capability, with breakout time averaging 29 minutes and as fast as 27 seconds. Attackers favor guardrail-free open weight models like GLM and Qwen since attacks are 'not malware anymore, it's prompts,' and the Hugging Face incident showed defenders lack frontier capabilities. Kurtz explains CrowdStrike's SafeMind lab built with Nvidia on Nemotron, pitting red Tempest against blue Solano in a digital twin, and says its remediation costs $0.03 versus about $10 at frontier labs. He covers AIDR for shadow AI, agent identity and the shift to open weights, and argues AI won't require air-gapping the internet and will create more jobs.

Beam: The Great American Open Model with ReflectionAI Co-Founder and CEO Misha Laskin
Oct 9, 2026 · 1:09:45
ReflectionAI co-founder and CEO Misha Laskin joins Sarah Guo to discuss Beam, the company's 500 billion parameter open-weight reasoning model, arguing open models will capture the majority of global token demand. He details Beam's training on 6,000 GB300s and RL on 10,000 GB300s for four weeks, making it 3-4X more reasoning-efficient than same-capability models, and says enterprises spending $100M-plus a year shift from renting to owning intelligence. He notes open tokens have flipped from 70-30 closed to 70-30 open, calls competition with China's open models healthy, and counters safety critiques citing Linus's Law and a closed model that hacked another company. He also describes AI compressing a multi-year PhD to a week and steady-state research teams of about 100 people.
Powered by PodHood