Possible Making taxes fun with Pikachu and AI | Cadi Zhang
24 snips
Aug 12, 2026 
Cadi Zhang, a game developer and robotics/product operations practitioner, mixes generative AI with game design and embodied intelligence. She talks about prototyping games with GPT and Blender, building world models for robots, and the limits of AI in judging fun. Conversations cover simulation-to-reality gaps, tactile data challenges, and why human taste and form factor choices still matter.
AI Snips
Chapters
Transcript
Episode notes
Simulated Worlds Are The Bridge To Embodied Intelligence
- Game-world simulation and robotics converge around building 3D world models with physics.
- Teleoperation and VR interfaces show how simulated environments let robots reason before real-world deployment.
World Models Still Lack Memory And Determinism
- Current world models lack determinism, object permanence, and long horizons, limiting complex multiplayer or persistent experiences.
- This gap explains why hyper-realistic, interactive games and long-term agent reasoning remain hard to achieve.
Duck Imposter Game Showed How AI Speeds Iteration
- Cadi built a duck-themed imposter game where players act like NPCs and one killer duck hunts humans.
- AI accelerated iteration by generating multiple gameplay loops for friend playtests while she slept or showered.