Vincony AI Glossary

A large language model (LLM) is a neural network trained on text (and often code, images, and audio) at massive scale to predict the next token in a sequence — giving it the ability to read, write, reason, summarize, and converse in natural language.

Multi-model AI is an architecture where one user can query multiple foundation models (e.g.

Retrieval-Augmented Generation (RAG) is an AI architecture pattern that combines a language model with an external knowledge source.

Model distillation is a technique for training a small, fast 'student' model to imitate a large, expensive 'teacher' model — transferring most of the teacher's capability into a fraction of the size.

Model routing is the automated decision of which AI model handles a given prompt.

A context window is the maximum amount of text (measured in tokens) that an LLM can consider in a single request — both the prompt the user sends and the answer the model generates.

AI memory is the ability of an AI assistant to remember context across sessions — facts the user has shared, preferences, ongoing projects, and previous conversations.

Agentic AI refers to AI systems that take multi-step actions toward a goal — choosing tools, calling APIs, browsing the web, writing files — rather than just answering a single prompt.

Multi-modal AI is a single model that can understand and/or generate more than one type of data — text, images, audio, and video — in one unified system.

Chain-of-thought is a technique where an AI model works through a problem step by step in writing before giving its final answer, mimicking how a person shows their working.

A token is the basic unit of text that an LLM reads and generates — typically 3-4 characters or roughly 0.75 of an English word.

Fine-tuning is the process of taking a pre-trained foundation model and continuing its training on a smaller, task-specific dataset so it specializes in a narrow domain, format, or style.

Inference is the act of running a trained AI model to produce an output — every time you send a prompt and get a response, that's one inference.

An embedding is a list of numbers (a vector) that represents the meaning of a piece of text, image, or audio in a way computers can compare mathematically.

Temperature is a setting (typically 0 to 2) that controls how random an AI model's output is.

Top-p, or nucleus sampling, is a setting (0 to 1) that limits an AI model's word choices to the smallest set of most-likely tokens whose combined probability reaches p.

A system prompt is a hidden instruction set given to an AI model before the conversation starts, defining its role, tone, rules, and boundaries.

Zero-shot means asking an AI model to perform a task with no examples in the prompt — just the instruction.

Few-shot means including a small number of worked examples in your prompt to show an AI model exactly what you want before asking it to do the task.

A tokenizer is the component that breaks text into tokens — the subword chunks an LLM actually reads and generates — and converts them to the numeric IDs the model processes.

Rate limiting is a cap a provider places on how many requests or tokens you can send in a given window — for example 500 requests per minute or 200,000 tokens per minute.

Latency is the delay between sending a prompt and getting the model's response.

A hallucination is an AI-generated statement that is plausible-sounding but factually wrong — an invented citation, a fictional API method, a wrong number stated with confidence.

Prompt injection is a security attack where malicious instructions are hidden inside content an AI processes — a web page, uploaded file, or email — tricking the model into ignoring its original instructions and following the attacker's instead.

A prompt library is a shared store of reusable AI prompts — typically with templates, variables, and access controls.

BYOK — 'bring your own key' — is a model where you connect your own provider API keys (OpenAI, Anthropic, Google) to a third-party app, so calls are billed to your provider account instead of the app's.

An API key is a secret string that authenticates your requests to a service's API, identifying who is calling and which account to bill.