AI Glossary & Learning Hub

The cost structure for using AI models through cloud APIs, typically charged per token processed. Input tokens and output tokens usually have different prices, with output tokens costing more.

The core component of transformer models that allows each token to dynamically focus on relevant parts of the input sequence. It computes weighted relationships between all tokens, enabling the model to understand context and dependencies.

Standardized tests used to measure and compare AI model performance across specific tasks like reasoning, coding, math, and language understanding.

A prompting technique that encourages the model to show its reasoning step by step before arriving at a final answer, significantly improving performance on complex reasoning, math, and logic tasks.

The maximum amount of text (measured in tokens) that a language model can process in a single request, including both the input prompt and the generated output.

A training technique where a smaller "student" model learns to replicate the behavior of a larger "teacher" model, producing a compact model that retains much of the teacher's capability at lower cost.

Dense numerical vector representations of text (or other data) that capture semantic meaning. Texts with similar meanings have similar embeddings, enabling search, clustering, and retrieval applications.

A prompting technique where you provide a small number of input-output examples within the prompt to guide the model's behavior on a new task, without any additional training.

The process of further training a pre-trained model on a smaller, task-specific dataset to improve its performance on particular tasks or to adapt it to a specific domain.

The most capable and advanced AI models available at any given time, typically produced by leading AI labs. They represent the current state of the art in performance across reasoning, coding, and general intelligence tasks.

A capability that allows AI models to generate structured outputs that invoke predefined functions or APIs, enabling models to take actions like searching databases, calling web services, or executing code.

Safety mechanisms and content filters implemented to prevent AI models from generating harmful, biased, or inappropriate outputs. These include both built-in model alignment and external validation systems.

When an AI model generates information that sounds plausible and confident but is factually incorrect, fabricated, or not grounded in the provided context. A major reliability challenge in AI applications.

The process of running a trained AI model to generate predictions or outputs from new inputs. This is what happens every time you send a prompt to an AI model and receive a response.

The time delay between sending a request to an AI model and receiving the response. Low latency is critical for real-time applications like chatbots and coding assistants.

A model architecture that uses multiple specialized sub-networks ("experts") and a routing mechanism that activates only a subset of them for each input, enabling larger total model capacity without proportional increases in computation.

AI models that can process and generate multiple types of data — such as text, images, audio, and video — rather than being limited to text alone.

AI models whose weights are publicly released, allowing anyone to download, run, modify, and fine-tune them. Examples include Meta's Llama, Mistral, and DeepSeek models.

The learned numerical values (weights) within a neural network that determine the model's behavior. More parameters generally means more capacity to store knowledge and handle complex tasks, but also higher computational costs.

The practice of designing and refining the text instructions (prompts) given to an AI model to elicit the most accurate, useful, and relevant responses for a given task.

A technique that reduces the precision of a model's numerical weights (e.g., from 16-bit to 4-bit), dramatically decreasing memory usage and increasing inference speed with only a small loss in output quality.

A technique that enhances AI model responses by retrieving relevant information from external knowledge sources and including it in the prompt, reducing hallucinations and enabling access to up-to-date or private data.

AI models specifically designed or trained to perform extended logical reasoning, mathematical problem-solving, and multi-step analysis. They typically use internal chain-of-thought processing before producing a final answer.

A training technique that uses human preferences to fine-tune AI models, making them more helpful, harmless, and honest. Humans rank model outputs, and the model learns to produce responses that align with those preferences.

A special instruction set provided to the model at the beginning of a conversation that defines its behavior, personality, constraints, and role. It persists across all messages in the conversation.

A parameter that controls the randomness of a model's output. Lower values (e.g., 0.0) make responses more deterministic and focused, while higher values (e.g., 1.0) make them more creative and varied.

The rate at which an AI model generates output, typically measured in tokens per second. Higher throughput means faster response generation and the ability to serve more concurrent users.

The basic units that language models use to read and generate text. A token can be a word, part of a word, or a punctuation mark — roughly 3/4 of a word in English on average.

A sampling method that limits token selection to the smallest set of tokens whose cumulative probability exceeds a threshold p. At top-p=0.9, the model considers only the top tokens that together account for 90% of the probability mass.

The neural network architecture that underpins virtually all modern large language models. Introduced in the 2017 paper "Attention Is All You Need," it processes text in parallel using self-attention mechanisms.

Asking a model to perform a task using only natural language instructions, without providing any examples. The model relies entirely on its pre-trained knowledge to understand and complete the task.