Duke Research: Simplicity from Chaos

The show leaned less on rapid breaking news and more on synthesis, reviewing Andrej Karpathy’s 2025 LLM year in review, practical experiences with Claude Code and Gemini, and what real human AI collaboration actually looks like in practice. The second half moved into policy tension around AI governance, advances in robotics and animatronics, autonomous vehicle failures, consumer facing AI agents, and new research on human AI synergy and theory of mind.

Key Points Discussed

Andrej Karpathy publishes a concise 2025 LLM year in review

Shift from RLHF to reinforcement learning from verifiable rewards

Jagged intelligence, not general intelligence, defines current models

Cursor and Claude Code emerge as a new local layer in the AI stack

Vibe coding becomes a mainstream development pattern

Gemini Nano Banana stands out as a major paradigm shift

Claude Code helps with local system tasks but makes critical date errors

Trust in AI agents requires constant human supervision

Gemini Flash criticized for hallucinating instead of flagging missing inputs

AI literacy and prompting skill matter more than raw model quality

Disney unveils advanced Olaf animatronic powered by AI and robotics

Cute, disarming robots may reshape public comfort with robotics

Unitree robots perform alongside humans in live dance shows

Waymo cars freeze in traffic after a centralized system failure

AI car buying agents negotiate vehicle purchases on behalf of users

Professional services like tax prep and law face deep AI disruption

Duke research shows AI can extract simple rules from complex systems

Human AI performance depends on interaction, not model alone

Theory of mind drives strong human AI collaboration

Showing AI reasoning improves alignment and trust

Pairing humans with AI boosts both high and low skill workers

Timestamps and Topics

00:00:00 👋 Opening, laptops, and AI assisted migration

00:06:30 🧠 Karpathy’s 2025 LLM year in review

00:14:40 🧩 Claude Code, Cursor, and local AI workflows

00:22:30 🍌 Nano Banana and image model limitations

00:29:10 📰 AI newsletters and information overload

00:36:00 ⚖️ Politico story on tech unease with David Sacks

00:45:20 🤖 Disney’s Olaf animatronic and AI robotics

00:55:10 🕺 Unitree robots in live performances

01:02:40 🚗 Waymo cars halt during power outage

01:08:20 🛒 AI powered car buying agents

01:14:50 📉 AI disruption in professional services

01:20:30 🔬 Duke research on AI finding simplicity in chaos

01:27:40 🧠 Human AI synergy and theory of mind research

01:36:10 ⚠️ Gemini Flash hallucination example

01:42:30 🔒 Trust, supervision, and co intelligence

01:47:50 🏁 Early wrap up and closing

The Daily AI Show Co Hosts: Beth Lyons and Andy Halliday