This is part of our beginner’s guide to using AI. AI coverage throws around terms like “context window” and “fine-tuning” constantly without explaining them, which makes the space feel more impenetrable than it needs to be. Here are 30 of the most common terms, explained without the jargon.
The basics
Large language model (LLM)
The type of AI system behind ChatGPT, Claude, and Gemini, trained on huge amounts of text to predict and generate language. “Large” refers to the size of the model and the amount of data it was trained on.
Prompt
The text you type to an AI assistant, your instruction or question. See our full guide to writing better prompts.
Chatbot
A general term for any program designed to hold a conversation with a user. Modern AI chatbots like ChatGPT are far more capable than earlier rule-based chatbots, but the term predates current AI by decades.
Token
The basic unit of text a language model processes, roughly a word or part of a word. Usage limits and pricing on many AI tools are measured in tokens rather than words or characters.
Context window
The maximum amount of text a model can consider at once in a single conversation, measured in tokens. Once a conversation exceeds this limit, the model starts losing track of the earliest parts.
How models behave
Hallucination
When an AI model states something false with the same confident tone as something true. Covered in depth in our guide to AI hallucination.
Temperature
A setting that controls how random or predictable a model’s output is. Lower temperature gives more consistent, conservative answers; higher temperature gives more varied, creative ones.
Hallucinated citation
A specific, common form of hallucination: a fabricated source, study, or reference that sounds real but doesn’t exist. Always verify a citation independently before using it.
Multimodal
A model that can process more than one type of input, text, images, audio, rather than text alone. Most leading assistants are now multimodal to some degree.
Agentic
Describes an AI system that can take multiple steps toward a goal on its own, planning, executing actions, checking results, rather than just responding to a single prompt. Covered in our guide to AI coding assistants, where this distinction matters a lot.
Training and technical terms
Training data
The text, images, or other content a model learned from before it was released. This determines what a model “knows” and can reflect biases present in that source material.
Fine-tuning
Additional, more specific training applied to an already-trained model to specialize it for a particular task or domain, rather than training a model from scratch.
Parameters
The internal values a model adjusts during training that determine its behavior. Often cited as a rough (and imperfect) proxy for a model’s size and capability.
Open-weight model
A model whose underlying parameters are published for anyone to download and run themselves, as opposed to only being accessible through a company’s hosted service. Covered in our guide to open source versus paid AI tools.
Inference
The process of a trained model actually generating a response to your prompt, as distinct from the training process that created the model in the first place.
Training cutoff
The date after which a model has no knowledge from its training data, since it was trained before that point. Tools with web search can access more recent information despite this cutoff.
Business and product terms
API
A way for developers to connect their own software directly to an AI model, rather than using it through a consumer-facing app like ChatGPT’s website.
Rate limit
A cap on how much you can use a tool in a given period, common on free tiers, that resets after a set amount of time.
Free tier
A version of a tool usable at no cost, typically with usage limits or access restricted to an older or less capable model version compared to paid tiers.
System prompt
Instructions given to an AI model before your own prompt, often set by the company or developer building on top of it, that shape its behavior and personality across a conversation.
Model version
A specific release of an AI model; providers regularly release new versions with different capabilities, and comparisons or claims about a tool can go stale as versions change.
Image and media generation terms
Diffusion model
The technical approach behind most AI image generators, including Stable Diffusion and Midjourney, which generates an image by gradually refining random noise into a coherent picture.
Upscaling
Increasing an image’s resolution after generation, either through the same tool or a separate one, to produce a sharper final version.
Text-to-image
The general category of AI tools, like DALL-E and Midjourney, that generate an image from a written description.
Voice cloning
Generating synthetic speech that mimics a specific real person’s voice, using a sample recording. Covered in our guide to AI voice generators, including the consent issues involved.
Text-to-speech (TTS)
Technology that converts written text into spoken audio, the core function behind AI voice generation tools.
Safety and ethics terms
Alignment
The general effort to make an AI model’s behavior match human intentions and values, rather than pursuing its literal training objective in ways that produce unintended or harmful results.
Guardrails
Restrictions built into an AI tool to prevent it from producing harmful, dangerous, or policy-violating content, even if a user specifically requests it.
Bias (in AI)
Systematic skew in a model’s output reflecting patterns, including unfair ones, present in its training data. Covered in more detail in our guide to AI ethics and privacy.
Jailbreak
An attempt to bypass a model’s built-in guardrails through specifically crafted prompts, generally against a tool’s terms of service and a genuine ongoing security concern for AI providers.
AI detector
A tool that attempts to identify whether text was AI-generated, with real accuracy limitations covered in our guide to how AI content detectors work.
Related reading
Back to the beginner’s guide to using AI, and our full AI tools directory to put these terms into practice.
Frequently asked questions
Do I need to know all these terms to use AI tools well?
No, most everyday use requires none of this vocabulary. It’s most useful for understanding AI coverage in the news, comparing tools more precisely, or having a technical conversation about AI with someone else.
Why do AI companies use so much jargon?
Much of it comes from the underlying computer science and machine learning research this technology is built on, where these terms have precise technical meanings, and it hasn’t always been translated well for a general audience as the technology moved into mainstream products.
Will these terms change as AI evolves?
Some will stay stable since they describe fundamental concepts (tokens, training data); others, especially product-specific or trendy terms, may fall out of use as the technology and its marketing language keep evolving.
[…] AI Glossary: 30 AI Terms Explained in Plain English […]