AI Glossary

AI systems designed to operate with autonomy — planning, reasoning, and executing multi-step tasks with minimal human guidance.

A hypothetical AI system that can understand, learn, and perform any intellectual task that a human can — unlike today's AI, which excels at specific tasks.

Technology that enables machines to perform tasks that typically require human intelligence, such as understanding language, recognizing images, and making decisions.

An AI system that can independently plan, make decisions, and take actions to accomplish goals — going beyond simple question-and-answer interactions.

The challenge of ensuring AI systems behave in ways that match human values and intentions, especially as they become more capable.

A set of rules that lets software applications communicate with each other — in AI, it's how developers integrate AI capabilities into their own products.

An AI agent that operates independently over extended periods, making decisions and taking actions without requiring human approval at each step.

Systematic errors in AI output that reflect prejudices in training data or design choices, leading to unfair or skewed results.

A prompting technique where you ask the AI to show its reasoning step by step — significantly improving accuracy on complex problems.

A software application that simulates conversation with users, ranging from simple rule-based systems to sophisticated AI-powered assistants like ChatGPT.

A saved snapshot of a trained AI model's weights at a specific point, commonly used in Stable Diffusion to produce different art styles.

Anthropic's approach to AI alignment where the model is trained to follow a set of explicit principles rather than relying solely on human feedback.

The maximum amount of text an AI model can process in a single conversation — measured in tokens. A larger context window means the AI can 'remember' more.

AI technology specifically designed for natural, human-like dialogue — the engine behind chatbots, voice assistants, and customer service automation.

A subset of machine learning using neural networks with many layers, enabling AI to learn complex patterns from large amounts of data.

AI-generated content that convincingly replaces a person's likeness or voice in media — raising serious concerns about misinformation and fraud.

The AI architecture behind most modern image generators — it works by learning to gradually remove noise from random static until a coherent image emerges.

A numerical representation of text (or images or audio) as a list of numbers, allowing AI to understand meaning and find similarities.

A specific URL or address where an AI API accepts requests — each capability (text generation, image creation, embeddings) typically has its own endpoint.

Providing a few examples in your prompt to show the AI the pattern you want — dramatically improving output quality for specific formats or styles.

The process of further training a pre-trained AI model on specific data to customize it for a particular task, domain, or style.

A large AI model trained on broad data that serves as the base for many applications — GPT-4, Claude, Gemini, and Llama are all foundation models.

An AI architecture where two neural networks compete — one generates content, one judges it — pushing each other to improve. Largely superseded by diffusion models for image generation.

A processor originally designed for rendering graphics that turned out to be ideal for AI training and inference due to its ability to perform many calculations simultaneously.

Techniques that anchor AI responses in verifiable facts and sources — reducing hallucination by connecting model outputs to real data.

Safety mechanisms built into AI systems to prevent harmful, inappropriate, or off-topic outputs.

When an AI confidently generates information that is factually incorrect, fabricated, or nonsensical — presenting it as truth.

The largest open-source AI platform — hosting models, datasets, and tools. Think of it as GitHub for AI models.

A system design where humans review, approve, or correct AI decisions at critical points rather than letting the AI operate fully autonomously.

The process of running a trained AI model to generate outputs — when you type a prompt and get a response, that's inference.

An AI technique that fills in selected areas of an image — used to remove objects, fix imperfections, or add new elements to specific regions.

Techniques used to bypass an AI model's safety guardrails and get it to produce outputs it was designed to refuse.

A lightweight data format used extensively in AI APIs to structure requests and responses between applications.

The date after which an AI model has no training data — it simply doesn't know about events, products, or information that emerged after this date.

The time delay between sending a request to an AI model and receiving the first response token — lower latency means faster, more responsive AI experiences.

A type of AI model trained on massive amounts of text data that can understand, generate, and reason about human language. GPT-4, Claude, Gemini, and Llama are all LLMs.

A lightweight fine-tuning technique that trains a small adapter on top of a base model, commonly used to add specific styles or characters to AI image generators.

Development platforms that minimize hand-coding through visual interfaces and pre-built components, while still allowing custom code when needed.

A branch of AI where systems learn patterns from data and improve through experience, rather than being explicitly programmed with rules.

An open protocol by Anthropic that standardizes how AI models connect to external tools, data sources, and services — like a universal adapter for AI capabilities.

An architecture where a model consists of multiple specialized sub-networks (experts), with a routing mechanism that activates only the relevant experts for each input.

A system where multiple AI agents work together, each with specialized roles, collaborating to accomplish complex tasks.

AI that can process and generate multiple types of content — text, images, audio, and video — rather than just one.

AI that is designed and trained for a specific task — as opposed to AGI, which would handle any intellectual task. All current AI is narrow AI.

The branch of AI focused on enabling computers to understand, interpret, and generate human language.

Instructions telling an AI image generator what NOT to include in the output — used to avoid common artifacts and unwanted elements.

A computing system inspired by the human brain, consisting of interconnected nodes (neurons) organized in layers that process information.

AI tools and platforms that let non-programmers build AI-powered applications, automations, and workflows without writing code.

A tool that makes running open-source AI models locally as easy as running a single command — no complex setup required.

AI models whose code and/or weights are publicly available, allowing anyone to use, modify, and distribute them — in contrast to proprietary models like GPT-4.

An AI technique that extends an image beyond its original borders, generating new content that seamlessly continues the existing scene.

The learned numerical values inside a neural network that determine its behavior — more parameters generally means more capability. GPT-4 is estimated to have 1.8 trillion parameters.

An AI-powered search engine that answers questions with cited sources — combining the conversational nature of ChatGPT with the citation rigor of traditional search.

The initial training phase where an AI model learns general knowledge from massive datasets before being specialized for specific tasks.

The text input you give to an AI model — your question, instruction, or creative brief that tells the AI what to do.

The practice of crafting effective prompts to get better results from AI models — a skill that combines clear communication, technical understanding, and creative problem-solving.

An attack where malicious text is embedded in user input to override the AI's system instructions or extract hidden prompts.

Reducing the precision of AI model weights (from 32-bit to 8-bit or 4-bit) to make models smaller and faster while sacrificing some quality.

A technique that improves AI accuracy by first retrieving relevant documents from a knowledge base, then using them as context when generating a response.

A restriction on how many AI API requests you can make within a time period — designed to manage server load and enforce usage tiers.

An AI model specifically designed to 'think through' complex problems step by step before answering — trading speed for accuracy on difficult tasks.

The practice of deliberately testing AI systems for vulnerabilities, biases, and failure modes by simulating adversarial use.

The process of finding and fetching relevant information from a knowledge base to provide context for AI generation — the 'R' in RAG.

A training technique where humans rate AI outputs, and the model learns to produce responses that humans prefer — the key method for making AI helpful and safe.

Structured data added to web pages that helps search engines understand the content — enabling rich results like star ratings, FAQs, and product info in search listings.

A collection of tools, libraries, and documentation that makes it easier for developers to integrate AI services into their applications.

Running AI models on your own hardware or private cloud instead of using a provider's API — giving you full control over data, costs, and customization.

Search that understands meaning and intent rather than just matching keywords — 'affordable places to eat nearby' finds budget restaurants even if they don't use the word 'affordable.'

AI that detects the emotional tone of text — identifying whether content is positive, negative, neutral, angry, sarcastic, or other emotions.

The leading open-source AI image generation model, created by Stability AI. Unlike Midjourney or DALL-E, it can be run locally on your own computer.

AI responses formatted in machine-readable structures (JSON, tables, lists) rather than free-form text — essential for building reliable AI applications.

A machine learning approach where the model learns from labeled examples — given inputs paired with correct outputs, it learns to predict the right output for new inputs.

Hidden instructions given to an AI model by the application developer that define the AI's behavior, personality, knowledge boundaries, and rules — invisible to the end user.

A setting that controls how random or creative an AI's responses are. Low temperature = predictable and focused. High temperature = diverse and creative.

AI technology that generates images from text descriptions — type what you want to see and the AI creates it.

AI that converts written text into natural-sounding spoken audio — used for voiceovers, audiobooks, accessibility, and content creation.

AI that generates video clips from text descriptions — one of the most rapidly advancing areas of AI in 2024-2025.

The basic unit of text that AI models process — roughly equivalent to a word or word fragment. AI pricing, context windows, and rate limits are all measured in tokens.

The process of splitting text into tokens — the fundamental preprocessing step before any AI language model can process your input.

The ability of an AI model to invoke external tools and APIs during a conversation — searching the web, running code, querying databases, or calling any programmatic function.

A generation parameter that controls which tokens the model considers — only tokens within the top probability mass are eligible, filtering out unlikely choices.

The process of feeding data to an AI model so it learns patterns and builds its capabilities — the foundation of all machine learning.

The data used to train an AI model — its quality, quantity, and composition directly determine the model's capabilities and biases.

Using knowledge a model learned from one task or dataset to improve performance on a different but related task — the principle behind fine-tuning.

The neural network architecture behind virtually all modern AI language models — introduced in 2017 by Google and now the foundation of GPT, Claude, Gemini, and Llama.

A specialized database optimized for storing and searching embedding vectors — the backbone of semantic search and RAG systems.

AI technologies that process and generate human speech — including text-to-speech, speech-to-text, voice cloning, and real-time voice conversation.

AI technology that creates a digital replica of a person's voice from a short audio sample, enabling text-to-speech in that voice.

The dedicated memory on a GPU used to store AI model data during processing. More VRAM = ability to run larger, more capable AI models locally.

An automated notification sent from one application to another when a specific event occurs — used to trigger AI workflows in response to real-time events.

OpenAI's open-source speech recognition model that converts spoken audio to text with high accuracy across 99 languages.

Tools that connect multiple apps and services to automatically perform sequences of tasks — enhanced by AI for intelligent decision-making and content generation.

An AI model's ability to perform a task it was never explicitly trained for — simply by understanding the task description.