AI Grounds
Open AI Grounds on a desktop
These interactive lessons need a larger screen. Please continue on a desktop or laptop computer.
AI Grounds
These interactive lessons need a larger screen. Please continue on a desktop or laptop computer.
AI Grounds
Start with the AI core, follow a specialist path, or explore an idea you need today. Learn by predicting, trying and explaining.
20 lessons · 0 coming soon
step 01
statistics
Move values on a number line, switch between dataset shapes, and see how the mean, median, and mode respond when values repeat or an outlier appears.
5 to 7 minutes
Updated
step 02
statistics
Move values on a number line, compare same-mean dataset shapes, and see how deviations, variance, and standard deviation react when points spread away from the mean.
5 to 7 minutes
Updated
step 03
inference
Repeatedly inspect reproducible samples from initially hidden populations and compare how sample size changes estimate variability without finite guarantees.
7 to 10 minutes
Updated
step 04
inference
Compare random samples with biased collection rules and see why size cannot fix bad sampling.
7 to 10 minutes
Updated
step 05
regression
Drag a regression line before revealing the least-squares best-fit line. Move slope and intercept across a loss surface to see why one line minimizes squared residuals.
Two short chapters
Updated
step 06
optimization
Tune learning rate and momentum while a point moves across a simple loss landscape.
5 to 7 minutes
Updated
step 07
generalization
Compare known-data fit with held-out prediction and expose data leakage.
6 to 9 minutes
Updated
step 08
generalization
Fit polynomial curves to noisy training dots, then compare them against held-out test dots.
5 to 7 minutes
Updated
step 09
classification
Move raw class scores before converting them into probabilities.
6 to 9 minutes
Updated
step 10
classification
Edit weights, bias and a decision cutoff on labeled 2D points to connect signed scores, sigmoid probabilities and classification boundaries.
7 to 10 minutes
Updated
step 11
probability
Edit four class logits and tune positive temperature to explore normalization, concentration, ranking and ties.
5 to 7 minutes
Updated
step 12
evaluation
Build confusion-matrix intuition before tuning a decision threshold or reading benchmark scores. Drag a classification threshold across scored examples.
Two short chapters
Updated
step 13
loss
Move probability mass onto and away from the true class to see confident mistakes get punished. Switch between binary, categorical, and multi-label cross entropy.
Two short chapters
Updated
step 14
optimization
Compare None, L1 and L2 on a fixed toy scorer, separate penalty selection from an explicit weight-grid fit, and track data loss, sparse weights, score ties and boundaries.
7 to 10 minutes
Updated
step 15
neural networks
Move input signals through ReLU, sigmoid, and tanh to see how neurons reshape values.
5 to 8 minutes
Updated
step 16
neural networks
Move weights in a tiny network and watch inputs become class scores.
6 to 9 minutes
Updated
step 17
neural networks
Trace three fixed formulas through forward caches and backward messages.
6 to 8 minutes
Updated
step 18
neural networks
Predict, reveal, and explain output-weight gradients, target changes, learning-rate scaling, and hidden-activation signals in one live computation graph.
5 to 7 minutes
Updated
step 19
neural networks
Step through forward values, batch loss, backward gradients and parameter updates, then compare repeated learning and held-out error.
9 to 12 minutes
Updated
step 20
neural networks
Inspect real WebGPU inference from input preprocessing to hidden neuron sums, softmax probabilities and local score saliency.
6 to 9 minutes
Updated