AI, Actually Explained · experiment

Teach a small model.

Label. Train. Test. Then change the examples and see what changes.

A simplified nearest-neighbour model. All calculations happen in this tab.

1. Label examples

0 labelled / 4 available

Call each drawn sample a leaf or a bloom. You can change any label.

Sample 01

Colour 0.08 Roundness 0.18

Sample 02

Colour 0.22 Roundness 0.25

Sample 03

Colour 0.78 Roundness 0.78

Sample 04

Colour 0.92 Roundness 0.84

Train with one example to start, or label several before training.

2. Test your model

8 separate samples

These sample IDs and measurements never enter the training set.

Optional. Write a prediction before testing, then compare it with the guesses. It stays in this tab.

After training, test what the model does with examples it has not stored.

Some leaves and blooms have different colours from the starter examples.

Label a sample to begin.

What this small model does

It compares colour and shape with the examples you labelled and stores those examples. Large neural models learn differently, by changing weights.

See the two measurements and the rule

Each illustration has hand-set measurements from 0 to 1. Colour runs from green to violet. Roundness runs from pointed to rounded. The model adds the squared differences in the two measurements, takes the square root, and uses the nearest example’s label. Equally close examples with conflicting labels produce “Cannot decide”.

This is an original toy collection of drawings. Its results describe these eight practice samples, and say nothing about accuracy on real plants or people.

Choices stay in this tab. Refresh starts over. This experiment awards no course credit.