96 terms
AI glossary
Saw a strange word in the news or in an app? One plain line for each, and a lesson when you want more.
96 terms
A
- A/B test
- Trying two versions of something with different groups of people to see which one works better.
- Agent
- An AI system that takes several steps on its own, such as searching, clicking or using apps, to finish a goal you set.
- Lesson: AI agents: AI that takes action →
- AGIArtificial General Intelligence
- A hoped-for or feared future AI that could learn and do almost any mental task as well as a person. It does not exist yet.
- AIArtificial Intelligence
- Computer software that learns patterns from examples so it can recognise, predict or produce things, instead of following only hand-written rules.
- Lesson: What is AI, really? →
- AI literacy
- Knowing enough about how AI works to use it well, spot its mistakes and decide when not to use it.
- Lesson: How to keep up without burning out →
- Algorithm
- A set of step-by-step instructions a computer follows to solve a problem or make a decision.
- Alignment
- The work of making AI systems do what people actually intend, and avoid harmful behaviour, even in unexpected situations.
- APIApplication Programming Interface
- A doorway that lets one program send requests to another, for example an app asking an AI model for an answer.
- Attention
- The part of a transformer that lets each word weigh how much every other word in the text matters to its meaning.
- Automation bias
- Our habit of trusting a machine's answer too readily, even when our own judgement or other evidence says otherwise.
- Lesson: Review AI like a manager →
B
- Benchmark
- A standard set of test questions used to score and compare AI models. High scores do not guarantee good results on your task.
- Lesson: Choosing a model →
- Bias
- When an AI treats some people or ideas unfairly because of skewed patterns in the data it learned from.
C
- Chain of thought
- Asking an AI to work through a problem in steps before answering, which often makes its answers more accurate.
- Lesson: Ask it to think step by step →
- Chatbot
- A program you talk to in everyday language by typing or speaking, which replies in the same way.
- Lesson: What a chatbot really does →
- Classifier
- A model that sorts things into groups, such as spam or not spam, or which plant is in a photo.
- Computer vision
- AI that makes sense of images and video, such as reading signs, spotting faces or checking products on a factory line.
- Content credentials
- Tamper-evident labels attached to a photo or video that record where it came from and whether AI was used to make or edit it.
- Lesson: Deepfakes and voice clones →
- Context engineering
- Choosing which documents, instructions and history to give an AI so it has exactly what it needs for a task.
- Context window
- The maximum amount of text an AI can look at in one go. Anything beyond it is forgotten for that conversation.
- Lesson: The context window: AI's short-term memory →
- Copilot
- An AI assistant built into a tool you already use, such as an email app or code editor, that suggests as you work.
D
- Data labelling
- People tagging examples, such as marking which photos show a cat, so a model can learn what each thing looks like.
- Data privacy
- Keeping personal or confidential information under control, including deciding what you type into AI tools and where it is stored.
- Lesson: What not to paste into AI →
- Deep learning
- Machine learning using neural networks with many layers. It powers most of today's speech, image and language AI.
- Deep research
- A chatbot mode that searches many web pages over several minutes and writes a longer report with sources.
- Deepfake
- A fake video, photo or audio clip made with AI to show a real person saying or doing something they never did.
- Lesson: Deepfakes and voice clones →
- Diffusion model
- A type of AI that makes images or video by starting from random noise and cleaning it up step by step.
- Distillation
- Training a smaller, cheaper model to copy the answers of a bigger one, keeping much of its skill.
E
- Edge AI
- AI that runs directly on a phone, camera or other device instead of on distant servers, so data can stay local.
- Embedding
- A list of numbers that captures what a piece of text or image means, so similar meanings get similar numbers.
- Lesson: Embeddings: meaning as a map →
- EvalEvaluation
- A test you design to check whether an AI system does your specific job well, using examples with known good answers.
- Explainability
- How well people can understand why an AI produced a particular answer or decision.
F
- Few-shot prompting
- Including a few examples of what you want in your prompt so the AI copies their style and format.
- Lesson: Show it examples →
- Fine-tuning
- Extra training of an existing model on your own examples so it gets better at one narrow kind of task.
- Lesson: Fine-tuning vs prompting →
- Foundation model
- A large general-purpose model trained on broad data that many different products are then built on top of.
G
- Generative AI
- AI that produces new text, images, audio, video or code, rather than only sorting or scoring existing things.
- GPUGraphics Processing Unit
- A chip that does huge numbers of small calculations at once. First built for video games, now the main engine for AI.
- Grounding
- Tying an AI's answer to real sources you supply or it looks up, so claims can be checked instead of invented.
- Lesson: Check what AI tells you →
- Guardrails
- Rules and filters around an AI system that block harmful, off-topic or unsafe requests and answers.
H
- Hallucination
- When an AI states something false with confidence, such as a made-up fact, quote or source.
- Lesson: Why AI makes things up →
- Human in the loop
- Designing a process so a person reviews or approves the AI's work before it has real consequences.
I
- Image generation
- Creating pictures from a written description, using AI trained on very large collections of images and captions.
- Inference
- The moment a trained model is actually used to produce an answer, as opposed to the earlier learning phase.
J
- Jailbreak
- A trick prompt designed to get an AI to ignore its safety rules and do something it normally refuses.
K
- Knowledge cutoff
- The date after which a model learned nothing new. Unless it can search the web, it may not know recent events.
- Lesson: Check what AI tells you →
L
- Latency
- The delay between asking an AI something and getting the start of its reply.
- LLMLarge Language Model
- An AI trained on huge amounts of text to predict the next word, which lets it write, summarise, translate and chat.
- Lesson: What a chatbot really does →
- Local model
- An AI model that runs on your own computer or phone, so your data does not leave the device.
M
- Machine learning
- Teaching computers by showing them many examples so they find the patterns themselves, instead of being given every rule.
- Lesson: How machines learn →
- MCPModel Context Protocol
- An open standard that lets AI assistants plug into apps and data sources, like a universal adapter for tools.
- Lesson: MCP: a universal plug for AI →
- Memory
- A chatbot feature that saves facts about you between conversations so you do not have to repeat them.
- Mixture of experts
- A model design where only some parts of the network switch on for each request, making large models faster and cheaper.
- Model
- The trained file of learned patterns that does the actual work inside an AI product. One model can power many apps.
- Lesson: Choosing a model →
- Multimodal
- Able to work with more than one kind of input or output, such as text, images, sound and video together.
N
- Neural network
- Layers of simple maths units, loosely inspired by brain cells, whose connections are adjusted during training to find patterns.
- Next-word prediction
- The core job of a language model: guessing the most likely next piece of text, over and over, to build a reply.
- Lesson: What a chatbot really does →
- NLPNatural Language Processing
- The field of computing that deals with understanding and producing human languages like English, Hindi or Swahili.
O
- OCROptical Character Recognition
- Turning text in a photo or scanned page into editable, searchable text.
- Open-source model
- A model whose trained weights are published so anyone can download, run and adapt it, often on their own hardware.
- Lesson: Choosing a model →
- Overfitting
- When a model memorises its training examples so closely that it does badly on new ones it has not seen.
P
- Parameters
- The adjustable numbers inside a model that are set during training. Big models have billions of them.
- Phishing
- Fake messages that pretend to be someone you trust to trick you into clicking a link, paying or sharing passwords.
- Lesson: Spot AI-powered scams →
- Pre-training
- The first, longest phase of building a model, where it learns general patterns from a huge mixed collection of data.
- Prompt
- The instruction or question you give an AI. Clearer prompts with context usually get much better answers.
- Lesson: The four-part prompt →
- Prompt chaining
- Breaking a big job into a sequence of smaller prompts, feeding each answer into the next step.
- Lesson: Prompt chaining: one step at a time →
- Prompt engineering
- The practice of wording, structuring and testing prompts to get reliable, useful results from AI.
- Lesson: The four-part prompt →
- Prompt injection
- Hidden instructions planted in a web page, email or file that try to hijack an AI reading it.
- Lesson: AI agents: AI that takes action →
Q
- Quantization
- Storing a model's numbers with less precision so it uses less memory and runs faster, with a small loss of quality.
R
- RAGRetrieval-Augmented Generation
- Having an AI look up relevant passages from your documents first, then answer using them, so replies are more accurate.
- Lesson: RAG: give AI your documents →
- Rate limit
- A cap on how many requests or messages you can send to an AI service in a set amount of time.
- Reasoning model
- A model that spends extra time working through a problem internally before replying, which helps with maths, logic and planning.
- Lesson: Choosing a model →
- Red teaming
- Deliberately attacking your own AI system, the way a bad actor would, to find weaknesses before release.
- Reinforcement learning
- Training by trial and error, where a system gets rewarded for good outcomes and learns to repeat what earned them.
- RLHFReinforcement Learning from Human Feedback
- Improving a model by having people rate its answers, then training it to give the kinds of answers people preferred.
- Role prompting
- Starting a prompt by telling the AI who to act as, such as an experienced editor, to shape tone and focus.
- Lesson: The four-part prompt →
S
- Semantic search
- Search that matches meaning rather than exact words, so "cheap flights" also finds "low-cost airfares".
- Lesson: Embeddings: meaning as a map →
- Small language model
- A compact language model that is cheaper and faster, often good enough for focused tasks and able to run on devices.
- Speech-to-text
- AI that turns spoken words into written text, as in meeting transcripts and voice typing.
- Lesson: Meetings and notes →
- Supervised learning
- Training a model on examples that come with the correct answer attached, like emails already marked as spam.
- Lesson: How machines learn →
- Sycophancy
- An AI's tendency to agree with you or flatter your ideas instead of telling you where you are wrong.
- Lesson: Check what AI tells you →
- Synthetic data
- Artificial examples made by computers, often by other AI, used to train or test models when real data is scarce or private.
- System prompt
- Background instructions set by the app maker, hidden from you, that tell the AI how to behave in every conversation.
- Lesson: What a chatbot really does →
T
- Temperature
- A setting that controls how predictable or varied an AI's word choices are. Low is steady, high is more surprising.
- Lesson: Temperature: safe or creative →
- Text-to-speech
- AI that reads written text aloud in a natural-sounding voice.
- Token
- A chunk of text, often part of a word, that AI reads and writes in. Limits and prices are counted in tokens.
- Lesson: Tokens: how AI reads →
- Tool use
- An AI's ability to call outside tools, such as a calculator, web search or calendar, to do things it cannot do alone.
- Lesson: AI agents: AI that takes action →
- Training
- The process of feeding a model many examples and adjusting its internal numbers until its outputs improve.
- Lesson: How machines learn →
- Training data
- The examples, such as text, images or recordings, that a model learns its patterns from.
- Lesson: How machines learn →
- Transformer
- The neural network design behind modern chatbots, good at tracking how every word in a passage relates to the others.
- Turing test
- An old thought experiment: if you cannot tell a machine from a person in conversation, does it think? Chatbots now often pass.
U
- Unsupervised learning
- Letting a model find groupings and structure in data on its own, without any answers provided.
V
- Vector database
- A database that stores embeddings so a system can quickly find the items closest in meaning to a question.
- Lesson: RAG: give AI your documents →
- Vibe coding
- Building software by describing what you want to an AI and accepting its code, often without reading it closely.
- Voice cloning
- Using AI to copy someone's voice from a short recording so it can say anything. Common in phone scams.
- Lesson: Spot AI-powered scams →
W
- Watermarking
- Hiding an invisible signal in AI-made text, images or audio so software can later detect where it came from.
- Lesson: Deepfakes and voice clones →
- Weights
- The learned numbers that make up a trained model. Sharing a model usually means sharing its weights file.
Z
- Zero-shot prompting
- Asking an AI to do a task with no examples, relying only on your instructions.
- Lesson: Show it examples →