AI glossary
AI Glossary in Plain English
This glossary explains common AI news terms without assuming you work in tech. Each definition is short, practical, and tied back to how the term appears in AI news or everyday tools.
For a guided path, use AI Explainers. For current examples, use Latest AI News.
Core terms
AI
AI means software that can do tasks that usually require human-like judgment, such as recognizing patterns, writing drafts, answering questions, or classifying information.
Example: A tool that summarizes customer reviews is using AI to find patterns in text.
Chatbot
A chatbot is a tool you talk to in messages. Some chatbots follow simple scripts, while newer AI chatbots can write, summarize, translate, and reason through instructions.
Example: ChatGPT, Claude, Gemini, and Copilot are AI chatbots.
Large language model / LLM
A large language model is an AI system trained on a very large amount of text so it can predict and generate language. It can answer questions, draft text, summarize documents, and help with code.
For current model news, see AI Models.
Generative AI
Generative AI creates new text, images, audio, video, code, or other content from a prompt. The output still needs review because it can be wrong or misleading.
Example: Asking a tool to draft a product description is a generative AI task.
Prompt
A prompt is the instruction or question you give an AI tool. Clear prompts usually include the task, audience, constraints, and desired format.
Example: “Summarize this email in five bullet points for a busy store owner.”
Hallucination
A hallucination is when an AI system gives an answer that sounds confident but is wrong, made up, or unsupported by the sources available to it.
This is why important AI answers should be checked against reliable sources.
AI agent
An AI agent is a system that can plan steps, use tools, and carry out a task with less hand-holding than a normal chatbot. It still needs boundaries and human review for important work.
Example: An agent might draft a support reply, look up an order, and suggest next steps.
Model
A model is the trained AI system behind a tool. The app is what you use; the model is the engine producing answers or predictions.
Example: Gemini, Claude, and GPT are model families used inside different products.
Training data
Training data is the information used to teach a model patterns before people use it. It can include text, images, code, audio, or other examples depending on the model.
Training data affects what a model is good at and where it may have blind spots.
Context window
A context window is how much information an AI model can keep in view at once during a conversation or task. Larger windows can handle longer documents, but they do not guarantee accuracy.
Example: A long context window may let a tool summarize a full report instead of a few paragraphs.
Multimodal AI
Multimodal AI can work with more than one kind of information, such as text, images, audio, video, or files.
Example: A multimodal tool might read a screenshot and explain what is shown.
Open-source AI
Open-source AI usually means some model code, weights, or related materials are shared so others can inspect, use, or adapt them. The exact license and limits matter.
For a recent example, see Britain Backs Open-Source AI.
API
An API is a way for one piece of software to use another service. In AI, developers use APIs to add model answers, summaries, image generation, or other AI features inside their own apps.
Example: A website might use an AI API to summarize support tickets.
RAG / retrieval-augmented generation
RAG is a method where an AI system retrieves outside information first, then uses that information while generating an answer. It can improve grounding, but the retrieved sources still need to be relevant and accurate.
Example: A company chatbot might retrieve help-center articles before answering a customer.
Benchmark
A benchmark is a test used to compare AI systems. Benchmarks can be useful, but they do not always show how a tool performs in ordinary real-world tasks.
Example: A model can score well on a test and still be awkward for normal office work.
Inference
Inference is the moment an AI model produces an answer or prediction after it has already been trained. It is the “using the model” stage.
Example: When a chatbot answers your prompt, it is doing inference.
Fine-tuning
Fine-tuning means training an existing model further so it becomes better at a specific task, style, or domain.
Example: A company might fine-tune a model on approved support replies.
Synthetic data
Synthetic data is data created artificially rather than collected directly from real events or people. It can help train or test systems, but quality and realism matter.
Example: A team might create sample customer questions to test a support chatbot.
Start with these definitions
New evergreen AI explainers
- AI agent: A plain-English guide to agents, tools, loops, and approval.
- RAG / retrieval-augmented generation: A simple guide to retrieval, grounding, and source limits.
- Training vs inference: A simple model-lifecycle distinction for AI news readers.
Flowchart
What RAG Does In Plain English
Retrieval-augmented generation is a way to give an AI answer extra approved context before it replies.
Flowchart showing a question, approved document search, retrieved snippets, AI answer drafting, caveats and sources, and human verification.
- User asks a question.The request starts with a normal prompt or business question.
- System searches approved documents or data.The tool looks in a selected source instead of relying only on memory.
- Relevant snippets are retrieved.Only the most related passages are passed forward as context.
- AI drafts an answer using the retrieved context.The model uses the prompt plus retrieved snippets to respond.
- Answer is shown with caveats/sources where available.Good systems show where important claims came from.
- Human verifies important claims.RAG can reduce unsupported answers, but it does not remove review.
Diagram
Chatbot Vs AI Agent
A chatbot usually replies. An agent can plan steps, use tools, observe results, and decide whether to continue.
Diagram showing an agent moving from goal to planning, tool action, observation, next step or stop, with human approval for risky actions.
- Goal or instructionA person gives the system a task, boundary, or outcome.
- Planning stepThe system breaks the goal into smaller next actions.
- Tool/action stepThe agent may call a tool, search, edit, calculate, or take another action.
- Observation/resultThe system checks what happened after the action.
- Next step or stopThe agent continues, asks for help, or ends the task.
- Human approval for risky actionsPurchases, publishing, deletions, and sensitive data use should require review.