Every term used in class, in plain language. Lesson tags show where you first meet each one.
Lesson tags (like 1-1-3) show where a term is first used, based on each lesson's topic. Your teacher may use a term in other places too.
The share of predictions a model gets right. One number can hide where it fails, which is why we also look at a confusion matrix.
An AI system that takes actions for you, like sending a message or changing a file, not just producing text. That makes permissions and guardrails matter more.
Your record of when you used AI, what you asked, what you checked, and how it went.
A set of step-by-step instructions a computer follows to get a result.
Computer systems that do tasks we usually think need human intelligence, like recognizing images, working with language, or making predictions. Today's AI mostly finds patterns in data. It does not think the way you do.
Using AI to make a person better at a task, rather than doing the task for them.
Having a machine do a task with little or no human effort.
A pattern where a system works better for some people or situations than others, often because of gaps or skews in the training data or in how the problem was framed.
A model that sorts an input into a category, like "cat or dog" or "spam or not spam".
A wrong answer delivered with complete certainty. Tone tells you nothing about whether an answer is correct.
A table showing, for each real category, what the model predicted. It shows which mistakes it makes, not just how many.
Someone's informed permission to use their data, work or likeness. Who gave consent, and whether they could refuse, matters for training data.
Everything the model can see when it answers: your prompt, earlier messages, and any pasted material. Better context usually means better answers.
Legal protection for original creative work. How it applies to AI training and AI output is still being decided in courts and policy, so check dates on what you read.
The trail of data your apps and devices collect about you.
Synthetic media that makes it look or sound like a real person said or did something they did not.
A running record of what you found, where it came from, and how sure you are. Your conclusions in Unit 3 come from it.
The model says no when the true answer is yes, like a spam email let through. Which error is worse depends on the situation.
The model says yes when the true answer is no, like a good email flagged as spam.
AI that creates new text, images, audio or video that resembles what it was trained on.
A rule or check built around an AI system to stop it doing harmful or unwanted things.
When an AI states something false or made up, such as a fake quote or source, in a fluent and confident way.
A design where a person reviews or approves what an AI system does before it takes effect.
Using a trained model to produce an answer for a new input. Training is learning; inference is using what was learned.
The correct answer attached to a training example, such as the word "cat" on a cat photo.
A model that predicts what text comes next. The dice model in Unit 1 is a tiny one.
A language model with billions of parameters trained on huge amounts of text. Chatbots are built on them.
Building a system by showing it many examples so it finds the pattern itself, instead of a person writing out every rule.
What you get after training: a set of numbers that turns an input into an output, like a photo into a label or a prompt into text.
A short, honest report about a model: what it was built for, what data it used, how well it works, and where it fails.
A model built from layers of simple connected units. Training adjusts how strongly they connect. It is loosely inspired by the brain but far simpler.
When a model memorizes its training examples instead of learning the general pattern, so it does well on training data and badly on new data.
One of the adjustable numbers inside a model. Training changes them. Large language models have billions; in Matchbox Nim the beads in each cup play the same role.
Changing what each person sees based on their profile. It can be useful and can also narrow what you are shown.
How likely something is, from 0 (never) to 1 (certain). A language model chooses each next word from probabilities, not from understanding.
A collection of data about what you do and like, used to personalize what you see and to predict your behavior.
The input you give an AI system: a question, instructions, examples, or all three.
Hiding instructions in text so an AI follows them instead of its real instructions. Anything an AI reads can carry them.
Where something came from and how it reached you. Knowing the provenance of a photo or claim helps you decide how far to trust it.
A model that predicts what you will want to watch, read or buy next, using your past behavior and other people's.
Deliberately attacking a system to find where it breaks, so the weaknesses can be fixed before real harm happens.
Learning by trying actions and getting rewards or penalties. Matchbox Nim does this with beads: moves that lead to wins get more beads.
The signal that tells a learning system an outcome was good. What you reward is what the system learns to chase.
A tool substitutes when it does a task instead of a person, and complements when it makes the person's work more valuable. The same AI can do both depending on the task.
A chatbot's habit of agreeing with you or praising your work even when you are wrong.
Images, audio or video created or changed by AI.
The mix of skills you have built that together make your work distinctive.
Breaking a job or problem into smaller tasks so you can look at each one separately.
The course method for breaking a job into tasks and judging, for each, whether AI replaces it, assists with it, or a human keeps it.
Examples the model has never seen, kept aside to check whether it really learned the pattern or just memorized.
A chunk of text, often a word or part of a word, that a language model reads and writes one at a time.
The process of adjusting a model's parameters using training data until it performs well.
The examples a model learns from. A model can only reflect patterns that are in its training data.
Being open about how a system works, what data it uses, and when AI was involved.
Checking an AI answer against something independent, like a primary source, your own calculation, or a person who knows.