How this works

1. We showed it examples. Using Google Teachable Machine, we recorded photos of a thumbs up and a thumbs down, each labelled. This is called supervised learning.
2. It found the patterns itself. Teachable Machine trained a model from those examples. Nobody wrote rules like "thumb pointing up means…".
3. It runs here, on your device. This page runs that model in your browser on your camera. Your video never leaves your device.
4. It only knows two things. Show no thumb, or a different face, and it will still pick one, often confidently. That's the point: models fail outside what they've seen.