AI Glossary

Software that performs tasks that normally require human intelligence — such as understanding language, recognising images, making decisions, or generating content. Modern AI does not 'think' like a human; it identifies patterns in data and uses them to produce outputs.

An AI system that can take a sequence of actions autonomously to complete a goal — browsing the web, writing and running code, sending messages — rather than just answering a single question. Claude can act as an agent when given the right tools.

A software program powered by artificial intelligence that can understand your questions and requests in plain language and respond helpfully. Claude, ChatGPT, and Gemini are all examples of AI assistants.

The AI safety company that built and maintains Claude. Founded in 2021 by former members of OpenAI, Anthropic focuses on building AI systems that are safe, reliable, and interpretable.

A set of rules that lets one piece of software talk to another. The Claude API lets developers send messages to Claude from their own applications — for example, to build a chatbot or automate document processing.

An AI assistant built by Anthropic. Claude is designed to be helpful, harmless, and honest. You interact with it by typing messages at claude.ai — no installation or technical knowledge required.

The maximum amount of text (your messages plus Claude's replies) that Claude can 'see' at once in a single conversation. Think of it as Claude's working memory — once the conversation exceeds this limit, earlier parts are no longer visible to Claude.

In Claude, a conversation is a single session where Claude remembers everything said so far. You can keep refining your request within the same conversation without starting over. Each new conversation starts fresh with no memory of previous ones.

A process where an AI model is further trained on a specific dataset to make it better at a particular task or domain — for example, training a model on medical literature to make it more accurate for clinical questions.

A large AI model trained on broad data that can be adapted for many different tasks. Claude, GPT-4, and Gemini are all foundation models — general-purpose systems that can write, reason, code, and converse.

AI that creates new content — text, images, audio, video, or code — rather than just classifying or analysing existing content. Claude is a generative AI: it generates new text in response to your prompts.

When an AI confidently states something that is factually incorrect. It is not intentional deception — it is a known limitation of how large language models are trained. Always verify specific facts, statistics, and citations from an independent source.

The process of running an AI model to generate a response. When you send a message to Claude and it replies, that reply is produced by inference — the model applying what it learned during training to your specific input.

The practice of refining Claude's output by asking follow-up questions or giving feedback within the same conversation. 'That is good but make it shorter' is an iteration. Skilled Claude users iterate rather than expecting perfection on the first try.

The date after which Claude was not given new information during training. Claude does not know about events, laws, products, or prices that changed after this date. You can ask Claude directly: 'What is your knowledge cutoff date?'

The type of AI technology Claude is built on — trained on vast amounts of text to understand and generate language. 'Large' refers to the enormous number of parameters (internal settings) the model uses to process and produce text.

In AI, a model is the trained system that processes your input and generates a response. Different Claude models (e.g. Claude 3.5 Sonnet, Claude 3 Opus) have different capabilities, speeds, and costs.

An AI model that can process more than one type of input — for example, both text and images. Claude is multimodal: you can paste an image into the conversation and ask Claude to describe, analyse, or answer questions about it.

The field of AI focused on enabling computers to understand, interpret, and generate human language. Claude is built on NLP technology — it understands your questions as natural sentences, not as formal commands.

The response or result Claude produces in reply to your prompt. 'Output' simply means what Claude writes back to you. The quality of the output depends heavily on the quality of the prompt.

The internal numerical settings of an AI model that are adjusted during training. A model with more parameters can generally handle more complex tasks — Claude's largest models have hundreds of billions of parameters.

The message you type to Claude — your instruction, question, or request. The quality of Claude's response depends heavily on how clearly and specifically you write your prompt.

The practice of crafting prompts carefully to get consistently excellent outputs from an AI model. It is not about writing code — it is about understanding how to communicate with AI effectively.

A technique that gives an AI model access to an external knowledge base at the time of answering — so it can retrieve relevant documents and use them to produce more accurate, up-to-date responses. Used to overcome the knowledge cutoff limitation.

A way of thinking about Claude's role: someone who helps you gather, summarise, and understand information — but whose work you review and verify before acting on it. Claude plays this role well, but is not a substitute for a qualified doctor, lawyer, or financial adviser.

A tool like Google or Bing that returns a list of links matching your query. Claude is different — it reads your full question and replies in plain language, like a knowledgeable person would, rather than returning a list of links to visit.

A set of instructions given to Claude before the user's conversation begins — used by developers to give Claude a specific role, personality, or set of rules. If you have ever used a customer service chatbot powered by Claude, its behaviour was shaped by a system prompt.

A setting that controls how creative or unpredictable Claude's responses are. Higher temperature = more varied, creative outputs. Lower temperature = more consistent, predictable outputs. Developers adjust this via the API; in the chat interface it is set automatically.

The unit Claude uses to process text — roughly 3–4 characters or about 0.75 words. Claude's context window and pricing are measured in tokens. As a rough guide: 1,000 tokens ≈ 750 words ≈ one to two pages of text.

The process by which an AI model learns — processing enormous amounts of text and adjusting its internal parameters to become better at understanding and generating language. Claude's training is what gives it its knowledge and capabilities.

The program you use to visit websites — for example Google Chrome, Safari, Firefox, or Microsoft Edge. You access Claude at claude.ai through a web browser. No additional software installation is required.

Using tools to connect repeatable tasks across apps and services so they work together with appropriate review and safeguards.