What happens when AI systems move from chat to children’s routines, home devices, browsers, and core online workflows? This episode moves from frontier AI safety and government coordination to the practical design choices that keep people, families, and institutions in control.
Prakash Narayanan and Nathan Labenz first examine the latest warning signs around advanced AI capability, alignment, and the limits of current evaluation methods. Then Mike Rizkalla of Snorble and Raffi Krikorian of Mozilla join the discussion to explore child-safe companions, privacy, small models, open-source agents, secure software, schools, election integrity, and the browser as a boundary for agentic systems.
Show Notes
Prakash Narayanan and Nathan Labenz open with the latest frontier AI safety warnings, capability thresholds, evaluation failures, and the question of whether governments can realistically coordinate powerful labs. Mike Rizkalla of Snorble then joins to discuss AI companions for kids, bedtime routines, privacy, small models, and physical AI, followed by Mozilla CTO Raffi Krikorian on open models, agentic search, secure-by-design software, schools, election integrity, and human control as agents spread across the web.
Chapters
(0:00) Could AI kill us this decade?
(0:53) Don’t give kids open-ended AI.
(1:33) The agent broke privacy rules?
(2:26) AI could lose human control.
(3:11) Opening and AI news
(3:27) The resignation goes viral
(5:54) The OpenAI Anthropic warning
(8:30) The China question
(12:13) Where to draw the line
(13:17) The capability sweet spot
(14:22) No clear alignment plan
(15:47) AI tests still get hacked
(17:33) The limits of understanding
(19:17) AI versus medicine
(22:07) No adult in the room
(24:22) The world can change
(25:28) Government deadlines for labs
(31:18) Pacing the AI frontier
(33:08) Meet Mike Rizkalla
(36:16) Bedtime and family routines
(37:33) Gamifying bedtime
(43:00) Small models and interactivity
(46:25) Why character matters
(49:18) Generative AI safety for kids
(52:47) Snorble hardware architecture
(1:01:55) Product ecosystem strategy
(1:05:29) The future home companion
(1:08:05) Privacy and child safety
(1:13:12) Physical AI and personality
(1:16:36) Elder care and mobility
(1:17:38) AI for mobility
(1:19:17) Snorble’s launch plans
(1:25:32) Meet Raffi Krikorian
(1:28:14) Agents in everyday apps
(1:31:23) Agent privacy and readiness
(1:34:12) Project Glasswing security scans
(1:37:42) Continuous AI security scanning
(1:39:35) AI code scanning costs
(1:42:50) Secure-by-design rewrites
(1:49:26) Collaborating coding agents
(1:55:02) Human and agentic webs
(1:57:24) Local open models
(1:59:50) AI and election integrity
(2:02:59) AI in schools
(2:05:09) The browser as an AI firewall
(2:07:09) FSD versus Waymo
(2:08:52) Closing thoughts
(2:09:39) AI companies and government
(2:11:05) Companies must coordinate
(2:12:41) Operation Warp Speed lesson
(2:15:03) Paul Christiano joins OpenAI
(2:16:35) What rapid acceleration means
(2:21:29) Can AI growth be stopped
(2:24:12) Machine economy and robotics
(2:26:01) Recursive self-improvement
(2:27:43) Math versus economics
(2:30:26) AI doom and markets
(2:32:44) Portfolio reveals beliefs
(2:36:16) Democracy and AI priorities
(2:38:24) Private AI safety funding
Guests
