The curator's reviewTrain your own image classifier in the browser in ninety seconds: show your webcam three poses, teach it your cat versus your other cat, no code, instantly downloadable.
Teachable Machine is Google's train-your-own-classifier playground: show your webcam examples of two or three classes - your face and your absence, two hand gestures, the cat versus the other cat - and it trains an image model in the browser, in seconds, and runs it live. Version 2 (2019) added audio and pose, plus model export, and quietly became the most-used ML education tool on Earth.
The pedagogical trick is speed. The gap between I have an idea and the machine does it is about ninety seconds, which is faster than the idea can die. Overfitting, bias, class imbalance - concepts that take lectures to convey - appear here as direct experience: train on photos taken only in daylight and watch your classifier fail at dusk, and now you understand training distributions forever.
Its fingerprints are all over a decade of delightful nonsense: science-fair projects, gesture-controlled everything, teachers building attendance systems from face classes, artists wiring sneeze detectors to disco lights. Half the weird things in this collection exist because someone's first model was two hand gestures and a houseplant, trained right here.
The gateway drug, distilled. Teach it to detect your I-am-about-to-procrastinate face; mine would never miss. Ninety seconds from idea to inference is a superpower, and this page has been handing it to children since 2017.
