Glossary.
Every term from the Learn AI course, defined in plain English and linked to the lesson that explains it.
- Agent
- A model running in a loop that plans, uses tools, checks results and continues until a goal is reached. Lesson 15
- AGI
- Artificial general intelligence, a system able to do roughly any intellectual task a person can; the term has no agreed definition or test. Lesson 19
- AI Act
- The European Union's broad AI law, which sorts uses by risk and places the strictest rules on systems that make decisions about people. Lesson 20
- AI winter
- A period, roughly the 1970s and again the late 1980s, when funding and interest in AI fell sharply after promises went unmet. Lesson 2
- API
- The application programming interface, the connection that lets a program send text to a model and receive text back. Lesson 16
- Artificial general intelligence (AGI)
- The idea of a system that could learn and do any intellectual task a person can; it does not currently exist. Lesson 1
- Artificial intelligence (AI)
- The broad field of making machines do tasks that look like they need intelligence when a person does them. Lesson 1
- Audience
- The person or group the answer is for, which sets its vocabulary, tone and depth. Lesson 9
- Base model
- The model as it comes out of pretraining: a strong text predictor that has not yet been trained to follow instructions. Lesson 6
- Benchmark
- A fixed set of test questions used to compare models, useful for researchers but narrower than real work. Lesson 19
- Brief
- A short statement of what you need, who it is for, what it should achieve and what tone it should take, given before any writing starts. Lesson 13
- Capability
- What a model can do at its best, as distinct from reliability, which is how often it does it. Lesson 19
- Citation
- A reference to a specific source, such as a paper, book, web page or court case, meant to let someone check a claim. Lesson 7
- Code interpreter
- A feature in some chat apps that lets the model write and run a small program, so arithmetic is calculated rather than guessed. Lesson 12
- Contamination
- When benchmark questions were in the training data, so a high score partly measures memory rather than skill. Lesson 19
- Content credentials
- A signed record attached to a file saying what created it and how it was edited. Lesson 14
- Context window
- The maximum amount of text, measured in tokens, that a model can take in at one time, including your messages, its replies and any files. Lesson 5
- Cross-check
- Confirming an answer against an independent source, such as another model, a search or the original document. Lesson 7
- Data curation
- Choosing, filtering and supplementing the material a model is trained on. Lesson 18
- Deepfake
- A synthetic image, audio clip or video that shows a real person doing or saying something they did not. Lesson 14
- Default
- What the model fills in when you leave something unspecified, usually the most common pattern in its training data. Lesson 18
- Delimiter
- A visible marker, such as a line of dashes or a pair of tags, that separates your instructions from pasted material. Lesson 10
- Demo
- A single impressive run of a capability, which shows what is possible but not how often it works. Lesson 19
- Diffusion model
- An image model trained to remove noise from a picture step by step, so it can build a new picture from pure static to match a description. Lesson 14
- Draft
- A first version of a reply that you expect to change, rather than a finished answer. Lesson 11
- Drift
- Details changing over a long task or conversation because earlier instructions and facts get crowded out. Lesson 8
- Edit and resend
- Changing an earlier message in the chat and sending it again, so the model starts from a clean brief instead of a pile of corrections. Lesson 11
- Edit pass
- A separate step where a draft is reviewed and fixed against a stated reader or goal, rather than written and sent in one go. Lesson 13
- Evaluation
- A structured test designed to measure how a model behaves, including whether its answers change when only a name or pronoun changes. Lesson 18
- Expert system
- A program from the 1970s and 1980s that stored a specialist's knowledge as a large set of hand-written if-then rules. Lesson 2
- Few-shot prompting
- Giving the model one or more examples of the output you want so it can copy the pattern. Lesson 10
- Fine-tuning
- Further training on a smaller, targeted set of examples, such as request-and-reply pairs, to change how the model behaves. Lesson 6
- Format
- The shape of the answer: a table, a list, a paragraph, an email, a set number of sentences. Lesson 9
- Fresh chat
- A new conversation with no history, used when the old thread has become noise. Lesson 11
- Frontier model
- The largest and most capable model a provider offers, usually also the slowest and most costly to run. Lesson 16
- Generative AI
- Machine learning systems that produce new text, images, audio, video or code rather than sorting or labeling what already exists. Lesson 1
- GPU
- A graphics card, hardware built to do millions of simple calculations at once, which made training large networks practical. Lesson 4
- Gradient descent
- The training method that nudges every parameter a little in the direction that reduces the network's error, over and over. Lesson 4
- Grounding
- Answering from specific text placed in front of the model rather than from its general training. Lesson 15
- Hallucination
- A fluent, confident answer from a model that is not true. Lesson 7
- Human in the loop
- A person checking and approving before an automated step that cannot be undone. Lesson 15
- Jagged profile
- The uneven pattern of a model's abilities: very strong at some tasks and weak at others that look similar. Lesson 8
- Knowledge cutoff
- The date after which no text was included in training, so the model knows nothing about later events on its own. Lesson 6
- Label
- The correct answer attached to one training example, such as "spam" or "not spam". Lesson 3
- Layer
- A group of neurons that all look at the same inputs; networks stack many layers, and "deep" means many of them. Lesson 4
- Limits
- Constraints on the answer, such as length, tone and things to leave out. Lesson 9
- Machine learning
- The part of AI where a system learns its behavior from examples in data instead of following rules written by hand. Lesson 1
- Model
- A function that turns an input into an output, with adjustable numbers inside that training sets. Lesson 3
- Multimodal
- Able to take in and produce more than one kind of material, such as text, images and audio, in the same model. Lesson 20
- Narrow AI
- A system that does one kind of task well and nothing else, which describes every AI system in use today. Lesson 1
- Negative prompt
- The part of an image request that says what to leave out, sometimes typed in a separate box. Lesson 14
- Neural network
- A learning system built from many simple connected units arranged in layers, loosely inspired by the brain. Lesson 2
- Neuron (artificial)
- A small unit that adds up its inputs, each multiplied by a weight, and passes a signal on if the total crosses a threshold. Lesson 4
- On-device model
- A model small enough to run on a phone or laptop without sending your data to a server. Lesson 20
- Open-weight model
- A model whose trained numbers are published so anyone can download and run it on their own hardware. Lesson 16
- Opt out
- A setting that excludes your conversations from being used to train future models. Lesson 17
- Overfitting
- When a model memorizes its training examples instead of learning the pattern, so it does well on them and badly on new data. Lesson 3
- Parameters
- All the adjustable numbers in a network, mainly its weights; modern language models have billions. Lesson 4
- Personal data
- Information that identifies or describes a specific person, such as a name tied to an address, health condition or ID number. Lesson 17
- Placeholder
- A stand-in such as [NAME] or [API_KEY] used in place of a real secret or private detail. Lesson 10
- Plausible
- Text that fits the pattern of what usually comes next, whether or not it happens to be true. Lesson 7
- Pretraining
- The first and longest stage of training, where the model learns by predicting the next token across a huge amount of text. Lesson 6
- Prompt
- The text you send to a model, including the instructions, any material and any examples. Lesson 9
- Prompt injection
- Text inside a page or document that is written to be read by the model as an instruction. Lesson 15 Lesson 17
- Prompt library
- Your own saved collection of prompts that worked, with a note on why each one did. Lesson 11 Lesson 20
- Reasoning mode
- A setting where the model works through intermediate steps before giving its answer, which helps on multi-step problems and slows down simple ones. Lesson 16
- Red flag
- A pattern in an answer, such as a precise number with no source, that signals it needs checking before use. Lesson 12
- Reinforcement learning
- Learning from rewards and penalties for actions rather than from correct answers. Lesson 3
- Reliability
- How consistently a model succeeds across repeated attempts, wordings and real data. Lesson 19
- Retention policy
- The provider's rule for how long it keeps your data before deleting it. Lesson 17
- Retrieval-augmented generation (RAG)
- Finding the most relevant passages in a set of documents first, then having the model answer from those passages. Lesson 15
- RLHF
- Reinforcement learning from human feedback: people rank the model's answers and the model is tuned toward the kind they preferred. Lesson 6
- Role
- A line that tells the model what kind of person to answer as, which shifts its vocabulary and focus. Lesson 10
- Sampling
- How a model picks each next token from several likely options with some randomness, so the same prompt can produce different answers. Lesson 8
- Second opinion
- Asking a different model, or a search engine, the same question to see whether independent sources agree. Lesson 12
- Self-critique
- Asking the model to judge its own reply against a stated goal before rewriting it. Lesson 11
- Self-supervised learning
- Learning by predicting a hidden part of the data, such as the next word in a sentence, so no human labels are needed. Lesson 3
- Server
- A computer run by the provider that receives your prompt, runs the model and stores the conversation. Lesson 17
- Stakes
- What happens if an answer turns out to be wrong; the thing that decides how much checking it deserves. Lesson 12
- Stereotype
- A generalization about a group of people that a model can absorb from how often words appear together in its training text. Lesson 18
- Synthesis
- A short piece that pulls several sources together into one account, with each claim tied back to where it came from. Lesson 13
- System prompt
- Standing instructions the app places before your first message, which shape how the model answers. Lesson 6
- Task
- The specific action you want, stated as a verb such as summarize, compare or draft. Lesson 9
- Temperature
- The setting that controls how much randomness goes into the model's choice of each next token. Lesson 8
- Test case
- A small check that runs code with a known input and confirms it gives the expected output. Lesson 13
- Text to speech
- Turning written text into a spoken voice, sometimes one cloned from a real person's recording. Lesson 14
- Token
- A chunk of text, often part of a word, that a language model treats as a single unit; in English about three quarters of a word on average. Lesson 5
- Tokenizer
- The step that chops text into tokens before the model sees it, using a fixed vocabulary of pieces. Lesson 5
- Tool use
- The model writing a request for the app to run something, such as a search or a calculation, and then reading the result. Lesson 15
- Trade-off
- What you give up in one option to gain something in another. Lesson 13
- Training
- The process of showing a model examples and adjusting its numbers a little each time to make its mistakes smaller. Lesson 3
- Training data
- The text, images or other material a model learns from, which for language models is mostly the public web plus books and other large collections. Lesson 18
- Transcription
- Turning recorded speech into written text, also called speech to text. Lesson 14
- Transformer
- The neural network design introduced in 2017 that uses attention to weigh how words relate to each other, and that underlies today's language models. Lesson 2
- Turing test
- Turing's 1950 proposal to judge a machine by whether its typed replies can be told apart from a person's. Lesson 2
- Turn
- One exchange in a chat, your message and the model's reply, after which the whole conversation is handed back to the model again. Lesson 5
- Two-factor authentication
- A second check at sign-in, usually a code from your phone, that protects an account even if the password leaks. Lesson 17
- Usage limit
- The cap a plan puts on how many messages, files or tokens you can use in a given period. Lesson 16
- Weak step
- The one step in a chain of reasoning that carries the real weight, and the one most worth checking. Lesson 12
- Weight
- The number attached to one input of a neuron that says how much that input counts and in which direction. Lesson 4
- Workflow
- A fixed sequence of prompts you run the same way each time for a certain kind of task. Lesson 13