no scary words here

AI Glossary

45 AI words in plain language, each linked to the lesson where you can try it for yourself.

45/45

Ask Bitu: The share of predictions a model gets right. It can hide poor results for small groups.; Try it in Where AI goes wrong
Ask Bitu: The curve a neuron passes its total through, such as sigmoid or ReLU. It lets networks learn non-straight patterns.; Try it in Weights and bias
Ask Bitu: Software that does tasks we link with human thinking, today mostly by learning patterns from data.; Try it in AI or not? The sorting game
Ask Bitu: A clear list of steps for solving a problem.; Try it in Rules vs learning

Ask Bitu: The method that sends error backwards through a network to work out how to adjust each weight.; Try it in Layers: going deep
Ask Bitu: When data or a model treats some groups unfairly, often because they were under-represented.; Try it in Where AI goes wrong
Ask Bitu: A number added to a neuron's total that makes it more or less eager to fire.; Try it in One neuron, three knobs

Ask Bitu: A program you talk to in text or speech. Modern ones use large language models.; Try it in AI in your day
Ask Bitu: Predicting a category, such as spam or safe.; Try it in Build a spam detector
Ask Bitu: Finding natural groups in data without any labels.; Try it in K-means clustering
Ask Bitu: Testing on several different splits of the data and averaging, for a steadier score.; Try it in Training vs testing, and overfitting

Ask Bitu: The examples a model learns from.; Try it in Data, model, feedback
Ask Bitu: An organised collection of examples.; Try it in Data, model, feedback
Ask Bitu: Machine learning with neural networks that have many layers.; Try it in Layers: going deep

Ask Bitu: One full pass through the training data.; Try it in Build a digit guesser

Ask Bitu: A number that describes an example, such as weight or length.; Try it in Features: how machines see data
Ask Bitu: A signal telling a model how right or wrong it was.; Try it in Data, model, feedback

Ask Bitu: Learning by repeatedly stepping downhill on the error.; Try it in How networks learn

Ask Bitu: When an AI confidently states something false or made up.; Try it in Where AI goes wrong
Ask Bitu: Neurons between a network's input and output, combining simple patterns into complex ones.; Try it in Layers: going deep

Ask Bitu: A clustering method that alternates assigning points and moving group centres.; Try it in K-means clustering

Ask Bitu: The right answer attached to a training example.; Try it in Build a spam detector
Ask Bitu: How big each training step is. Too big blows up, and too small is slow.; Try it in How networks learn
Ask Bitu: A small valley in the error that isn't the lowest point overall.; Try it in How networks learn
Ask Bitu: A number measuring how wrong a model is. Training tries to lower it.; Try it in How networks learn

Ask Bitu: Teaching computers from examples instead of writing every rule.; Try it in Build a spam detector
Ask Bitu: What's left after training: the thing that makes predictions.; Try it in Data, model, feedback

Ask Bitu: A simple classifier that combines how often each word appears in each class.; Try it in Build a spam detector
Ask Bitu: AI built for one specific task. All of today's everyday AI tools are narrow.; Try it in AI in your day
Ask Bitu: A model made of connected simple units (neurons) arranged in layers.; Try it in One neuron, three knobs
Ask Bitu: A small unit that multiplies inputs by weights, adds them up and decides.; Try it in One neuron, three knobs

Ask Bitu: An example very different from the rest, which can drag a model off course.; Try it in Linear regression
Ask Bitu: Memorising the training data so well that new data is handled badly.; Try it in Training vs testing, and overfitting

Ask Bitu: An early single-neuron learning model, built by Frank Rosenblatt in 1958.; Try it in One neuron, three knobs
Ask Bitu: One tiny dot of an image. To a computer, it's just a number.; Try it in Build a digit guesser
Ask Bitu: A model's output for a new example.; Try it in AI in your day

Ask Bitu: Predicting a number, such as rent or marks.; Try it in Linear regression

Ask Bitu: An S-shaped curve that squashes any number to between 0 and 1.; Try it in Weights and bias
Ask Bitu: Learning from examples that come with the right answers.; Try it in Build a spam detector

Ask Bitu: Examples hidden during training and used for an honest score.; Try it in Training vs testing, and overfitting
Ask Bitu: Adjusting a model's numbers to make fewer mistakes on examples.; Try it in Weights and bias

Ask Bitu: A model too simple to capture the pattern, even on training data.; Try it in Training vs testing, and overfitting
Ask Bitu: Finding structure in data without any right answers given.; Try it in K-means clustering

Ask Bitu: A number that sets how much an input counts, and in which direction.; Try it in Weights and bias

Ask Bitu: “Exactly one is on”. A pattern a single neuron can't learn but a hidden layer can.; Try it in Layers: going deep