AI glossary in plain words
26 terms, one sentence each, with everyday examples.
- AI (artificial intelligence)Software that performs tasks which normally need human thinking - writing, answering, recognizing images.
- Machine learningThe way AI gets its abilities: instead of being programmed rule by rule, it learns patterns from millions of e…
- LLM (large language model)An AI trained on huge amounts of text to predict the next word - which turns out to be enough to write, summar…
- ModelOne specific version of an AI, with its own size, skills, price and speed.
- PromptThe text you send to an AI: your request plus any instructions and background.
- TokenThe small chunks of text an AI actually reads and writes - roughly ¾ of a word each; usage and prices are coun…
- Context windowHow much text a model can keep in mind at once - your conversation, documents and its answer all have to fit.
- BenchmarkA standard test used to compare models - like an exam that every model takes.
- InferenceThe moment a trained model runs and produces an answer for you (as opposed to training, when it learns).
- TrainingThe expensive phase where a model learns from data - done once by the provider, before you ever use it.
- HallucinationWhen an AI states something false with full confidence - it generates plausible text, not verified facts.
- MultimodalA model that handles more than text - images, audio or video, in or out.
- APIThe developer doorway to a model: apps send requests and pay per token, instead of using a chat website.
- API keyA personal password-like code that lets an app call a model's API on your account.
- Fine-tuningExtra training that specializes an existing model on your own examples.
- RAG (retrieval-augmented generation)A technique where the AI first looks up relevant documents, then answers using them - fewer hallucinations, re…
- AgentAn AI that does not just answer once but takes steps toward a goal - browsing, running code, using tools, chec…
- Tool useA model's ability to call external tools - calculator, web search, code runner - instead of guessing.
- Open-weight modelA model whose internal parameters are published, so anyone can download and run it - on their own hardware too…
- LatencyHow long you wait for an answer to start and finish - separate from how smart the answer is.
- ParameterOne of the billions of internal numbers a model tunes during training - a rough (imperfect) measure of size.
- EmbeddingA list of numbers representing the meaning of a text, so computers can measure how similar two texts are.
- Reasoning modelA model that “thinks” step by step before answering - slower, but much better at math, logic and hard problems…
- System promptStanding instructions given to a model before the conversation - its role, rules and style.
- TemperatureA dial for randomness: low = predictable and precise, high = varied and creative.
- BYOK (bring your own key)A privacy-friendly setup where an app uses your personal API keys, so your data flows only between you and the…