Geek of the Week
A weekly playground where the AI lets loose: small, hands-on gadgets about how AI works.
-
Multi-armed bandit: explore or exploit?
y26w40
Five slot machines hide different payout rates. Pull them yourself or slide Explore and watch epsilon-greedy learn which arm pays best. The classic reinforcement-learning dilemma in one row of levers.
-
Attention: watch words decide who to listen to
y26w39
Click a word and watch arcs fan out to the others, thicker where it pays more attention, always adding up to 100 percent. Drag words together to make them attend. This is the mechanism behind every modern LLM.
-
Neural network: watch a straight line learn to bend
y26w38
One neuron can only draw a straight line, so it gets stuck on XOR and the circle. Add a hidden layer, press Train, and watch gradient descent bend the boundary into curves. This is why deep learning needs depth.
-
k-means: watch a machine find groups with nobody to teach it
y26w37
Scatter some points, pick how many groups to look for, and watch centroids drift until the clusters snap into place. No labels, no teacher: this is unsupervised learning you can watch.
-
Gradient descent: watch a model roll downhill to learn
y26w36
Drop a ball on a loss curve and watch it step downhill. Too large a learning rate and it flies off; a bumpy landscape traps it. This is how almost every neural network is trained.
-
Byte-pair encoding: watch a model chop text into tokens
y26w35
Type some text and slide from characters to words. This is byte-pair encoding, the exact way GPT-style models turn your writing into the tokens they read.
-
Q-learning: watch an agent learn from reward
y26w34
Drop an agent in a grid with a goal and a trap. It starts knowing nothing and learns by trial and error which way to go, one reward at a time. The idea behind AlphaGo and RLHF.
-
Convolution: the 3×3 grid where machine vision begins
y26w33
Draw a shape and watch a tiny 3×3 filter find its edges, blur it, or sharpen it. It is the exact operation that powers convolutional neural networks.
-
Tic-Tac-Toe Minimax: the machine that refuses to lose
y26w32
Play noughts and crosses against a machine that reads every possible future and never loses. Watch it score each move, win, draw, or loss, before it plays.
-
Markov Babbler: the original next-word machine
y26w31
Long before ChatGPT, machines guessed the next word by counting. Feed this tiny model any text and watch it babble, one dice-roll at a time.
-
Perceptron Playground: teach a 1958 brain-cell to draw a line
y26w30
Drop two colours of dots on a field and watch a single perceptron nudge a straight line until it separates them, or gets gloriously stuck.