Glossary
The AI glossary
Clear, neutral definitions of the AI and privacy terms that show up across SearchQ, so you always know exactly what a word means.
- AI chat
- A conversational interface where a person exchanges natural-language messages with an artificial-intelligence model that generates written replies in real time.
- AI council / model ensemble
- A technique that sends the same prompt to several AI models at once and combines their answers into a single synthesized response. Pooling multiple models tends to be more reliable than relying on any one of them alone.
- Confidential computing / secure enclave
- A hardware-isolated execution environment that keeps data encrypted in memory while it is being processed, so that even the host operating system or cloud provider cannot read it.
- Context window
- The maximum amount of text, measured in tokens, that a language model can consider at one time. It covers both the input prompt and the generated output, so longer conversations or documents can exceed it.
- End-to-end encryption
- A method of securing data so that it is encrypted on the sender’s device and can be decrypted only by the intended recipient, leaving any intermediaries unable to read it in transit.
- Fine-tuning
- The process of further training an already pre-trained model on a smaller, task-specific dataset to specialize its behavior or knowledge for a particular use case.
- Hallucination
- An output in which an AI model presents fabricated or factually incorrect information as if it were true. Hallucinations happen because models predict plausible text rather than retrieve verified facts.
- Incognito mode
- A chat setting that avoids keeping a lasting record of the conversation, typically deleting it after the session so no long-term history is stored.
- Inline verification
- A fact-checking step in which a separate model reviews an answer and marks individual claims as supported, questionable, or unsupported directly within the response text.
- Large language model (LLM)
- A neural network trained on very large amounts of text to predict and generate human-like language. LLMs power tasks such as answering questions, writing, summarizing, and translating.
- Model routing
- The automatic selection of which AI model should handle a given request, based on factors such as the question’s difficulty, the required speed, and cost.
- On-device AI
- AI inference that runs locally on a user’s own device instead of on a remote server, so the data being processed never has to leave the device.
- Prompt
- The input text a person or system gives an AI model to elicit a response. The wording, context, and instructions in a prompt strongly influence the quality of the output.
- RAG (retrieval-augmented generation)
- A technique that first retrieves relevant documents or data and supplies them to a language model as context, so the answer is grounded in specific, up-to-date sources rather than the model’s memory alone.
- Token
- The basic unit of text an AI model processes, usually a word or word fragment. Models measure input and output length, and often pricing, in tokens.
- Web search grounding
- Connecting a language model to live web search so it can base its answer on current information and cite the sources it used, rather than relying only on what it learned during training.
- WebGPU / WASM inference
- Browser technologies, WebGPU for GPU computation and WebAssembly (WASM) for near-native CPU code, let AI models run directly inside a web page, without sending data to a server.
- Zero data retention (ZDR)
- A processing policy under which a provider does not store or retain a user’s inputs or outputs once a request has been completed.