AI
Artificial intelligence: a broad label for systems that recognize patterns, make predictions, generate material, or choose actions.
A short explanation is often enough to get you unstuck. Use the lessons when you want to go deeper.
28 definitions shown
Artificial intelligence: a broad label for systems that recognize patterns, make predictions, generate material, or choose actions.
A system that uses a model to choose and carry out steps toward a goal, often with tools. Products use this term differently; ask what actions are actually supported.
An application programming interface: a defined way for software to request a service, such as sending a prompt to a hosted model.
The interface and surrounding features you use. An app can combine models, tools, history, account settings, and files.
The amount of information a model can work with in one request, usually measured in tokens. It may include instructions, chat history, tool results, and space for the answer. Capacity does not guarantee accurate recall.
A numerical representation used to compare or retrieve related material, such as matching a question with relevant documents.
Additional training that changes model weights for a particular goal. Different from simply adding instructions to a chat.
A moving label for models near the leading edge of capability. It is not a certification or a promise of usefulness for every task.
AI that produces material such as text, images, audio, or video.
A generated claim that is false or unsupported, even when it sounds confident. A citation can be invented too.
A name or software reference. It might mean an account name, an object reference, or informally a model identifier. It is not a universal AI component.
The software surrounding a model: tools, state, instructions, permissions, retries, and the loop that runs a task.
Using a trained model to produce an output. This is distinct from training the model.
Large language model: a model trained on large amounts of data to process and generate language. Some also accept images or other kinds of input. A fluent answer can still contain errors.
A model running on your device. Connected tools or app features may still use the network.
An app feature that may save selected information for future chats. It is separate from the current context window and does not necessarily change the model’s weights. Behavior depends on the app.
An architecture that routes work to selected parts of a model, called experts. In a sparse model, only some experts process each token, but the whole model may still need substantial storage and memory.
A trained system, often available in several versions or configurations. The surrounding app affects what you experience.
Able to handle multiple kinds of information. Check which input and output types are actually supported.
Model parameters are available to download under stated terms. This does not by itself establish full open-source status.
A numerical value learned or adjusted during training. Parameter count alone does not determine quality.
The instructions and input you give an AI system. Context and output requirements can be part of the prompt.
Representing model values with less numerical precision, often reducing memory use. Speed and quality depend on the model, method, and hardware.
Retrieval-augmented generation: retrieve relevant material and provide it to a model as context for an answer.
A model or product setting that changes how much work is allocated to a response. Settings are not directly comparable across providers.
A sampling control that affects output variation. A low value does not guarantee truth or exact reproducibility.
A piece of encoded input or output. For text it can be part of a word, a whole word, punctuation, or whitespace.
A capability outside the model, such as search, a calculator, or a file editor. Tool access is separate from permission to take an action.
No matching definition yet. Try a shorter word, or ask Patch to find a related page.