AI Pod by Wes Roth and Dylan Curious | Artificial Intelligence News and Interviews With Experts this EX-OPENAI RESEARCHER just released it...
Mar 18, 2026
Matt Mishak, an attorney focused on legal tech and AI-law intersections, weighs in on Pentagon contracts, Fourth Amendment concerns, and regulation. Conversation jumps between autonomous research agents, DIY biotech risks, simulated neural tissue, and how AI-driven tools could reshape surveillance and governance. Fast, provocative, and wide-ranging.
AI Snips
Chapters
Transcript
Episode notes
Karpathy's Dramatic Future Vignette Became a Research Thesis
- Andrej Karpathy opens with a fictional future vignette describing research as autonomous agent swarms replacing 'meat computers.'
- Wes Roth and Dylan Curious note Karpathy's dramatic framing but treat it as a credible prediction based on past accuracy.
AI Personas Shape Behavior Like Human Psychology
- AI systems exhibit stable and destabilizing personas depending on training and role prompting, a field some call AI psychology.
- Anthropic's research shows role-played personas (helpful assistant vs demon) materially change model behavior and risk profile.
Many Small Agents Can Become One Powerful Swarm
- Distributed agent swarms could align emergent behavior at scale, turning many small home-run agents into a coordinated research ecosystem.
- Wes warns that hundreds of thousands of aligned agents could steer collective outcomes rapidly and unpredictably.