Keep this page open while reading AI news. Each term gets a one-line definition; our full AI Glossary has longer entries.
- Model — a program that has learned patterns from data and uses them to make predictions or generate content.
- Parameters — the internal numbers a model adjusts during training; “70B” means 70 billion of them.
- Training — the process of adjusting parameters by showing the model many examples.
- Inference — using a trained model to produce an output.
- Token — a chunk of text (often part of a word) that language models read and write.
- Context window — how many tokens a model can consider at once.
- Large language model (LLM) — a model trained on huge amounts of text to predict and generate language.
- Transformer — the neural network architecture behind most modern LLMs.
- Prompt — the instructions and input you give a model.
- Hallucination — a confident but false output.
- Fine-tuning — extra training on a smaller dataset to specialize a model.
- RAG (retrieval-augmented generation) — giving a model relevant documents at question time so it can answer from them.
- Embedding — a list of numbers representing the meaning of text or images.
- Multimodal — able to handle more than one type of data, such as text and images.
- Open-weight model — a model whose trained parameters are published for anyone to download.
- Mixture of experts — an architecture where only part of the model is used for each token, saving compute.
- Agent — an AI system that can take actions, such as calling tools or browsing, to reach a goal.
- Benchmark — a standard test used to compare models.
- Alignment — work to make AI systems behave according to human intentions and values.
- GPU — the type of processor most AI training and inference runs on.
Terms you’ll see in AI product news
- Reasoning model — a model trained to work through problems step by step before answering.
- Inference cost — the price of running a model, usually charged per million tokens.
- Latency — how long a model takes to respond.
- System prompt — hidden instructions that set a model’s behavior in an app.
- Guardrails — rules and filters that limit what an AI system can say or do.
- Prompt injection — an attack that hides instructions in content an AI reads.
- Distillation — training a smaller model to imitate a larger one.
- Quantization — storing model numbers with less precision so it runs on smaller hardware.
- Synthetic data — training data generated by AI models.
- AGI (artificial general intelligence) — a hypothetical AI matching or exceeding humans across most tasks; definitions vary.
How to learn new terms quickly
When you meet an unfamiliar term, ask: what problem does it solve, and what’s the simplest example? Look it up in our AI Glossary, or ask an AI assistant to explain it at a beginner level with an analogy — then check one reliable source. Keeping a personal list of terms with your own one-line definitions helps them stick.
Terms that are often misused
- “AI” is used for everything from simple rules to frontier models.
- “Open source” is often used for models that only release weights under restrictive licenses.
- “Hallucination” describes confident errors — not imagination in a human sense.
- “Understands” and “thinks” are convenient shorthand, but their meaning for AI is debated.
What to watch next
- New vocabulary from agents and reasoning models.
- Standard definitions in regulation, such as the EU AI Act.
- Debates over terms like “open source AI” and “AGI”.
- Glossaries published by standards bodies.
Key takeaways
- A few dozen terms cover most of what you read in AI news.
- Tokens, context windows and parameters describe model size and capacity.
- Fine-tuning, RAG and embeddings describe how models are adapted and connected to data.
- Agents, guardrails and prompt injection matter as AI takes actions.
- Some terms are used loosely; check what is meant in each context.
Frequently asked questions
What is the difference between a model and an app?
The model is the trained AI system; an app like a chatbot wraps it with an interface, instructions and features.
What does “70B” mean?
Seventy billion parameters — the internal numbers the model learned during training.
Where can I find more definitions?
Our AI Glossary has longer explanations with examples and links.