24:22
Tina Huang
Every Size Local AI In 24 Minutes
Tina Huang lines up every computer she owns, sorts it by memory class, and tries to run an AI model on every rung: a 32 kilobyte Arduino that fits nothing, an 8 megabyte ESP32-S3 running TinyStories, a Raspberry Pi 5 chaining Whisper, Qwen3 1.7B and Piper into a talking assistant, an iPhone 16 with the same 8 gigabytes that behaves nothing alike, a 32 gigabyte MacBook Pro, a 64 gigabyte Mac Studio, a 128 gigabyte AMD Ryzen AI Halo, and rented GPUs up to an eight way H100 node with 640 gigabytes. A restaurant kitchen analogy explains why capacity and bandwidth are separate questions, and a one line formula tells you exactly what fits on your own machine.