AI Glossary: 20 Essential Terms Everyone Should Know
Updated July 20, 2026 · Reference Guide
The AI field has developed its own vocabulary, and understanding these terms is essential for evaluating tools, reading documentation, and communicating effectively about AI. This glossary covers the 20 terms that appear most frequently in AI tool documentation, reviews, and discussions — written in plain language without assuming technical background.
Whether you're reading a product comparison, evaluating an AI tool for your team, or just trying to understand the latest AI news, these terms will help you navigate the conversation with confidence.
AGI (Artificial General Intelligence)
A theoretical AI system that can understand, learn, and apply knowledge across any domain at a human level or above. No current AI system qualifies as AGI. The term is used to distinguish current narrow AI (designed for specific tasks) from a future system with general cognitive capabilities.
API (Application Programming Interface)
A set of protocols that allows different software applications to communicate with each other. In the AI context, APIs let developers integrate AI capabilities (like text generation or image recognition) into their own applications without building the models from scratch.
Chain of Thought
A prompting technique where the AI is asked to show its reasoning step by step before arriving at a conclusion. This improves accuracy on complex reasoning tasks because it forces the model to process intermediate steps rather than jumping directly to an answer.
Context Window
The amount of text (measured in tokens) that an AI model can process in a single conversation. Larger context windows allow the model to analyze longer documents and maintain more conversation history. Claude's 200K token window can process a 500-page book; smaller windows may struggle with long articles.
Embedding
A mathematical representation of text (or images, audio) as vectors in a high-dimensional space. Embeddings capture semantic meaning, allowing AI systems to measure similarity between concepts. They power search, recommendation, and clustering applications.
Fine-Tuning
The process of further training a pre-trained AI model on a specific dataset to specialize it for a particular task or domain. Fine-tuning a general language model on medical literature, for example, produces a model that performs better on medical questions.
Generative AI
AI systems that create new content — text, images, audio, video, or code — rather than simply classifying or analyzing existing data. ChatGPT, Midjourney, and DALL-E are all generative AI tools. This distinguishes them from analytical AI systems that make predictions or classifications.
Hallucination
When an AI model generates information that is false or fabricated but presents it confidently as factual. Hallucinations occur because language models predict likely sequences of words without verifying factual accuracy. They are the primary reason AI output requires human verification.
Inference
The process of using a trained AI model to generate output. When you type a prompt into ChatGPT and receive a response, that's inference. Inference requires computational resources, which is why AI tools have usage limits and pricing tiers.
Large Language Model (LLM)
A type of AI model trained on massive amounts of text data to understand and generate human language. GPT-4, Claude, and Gemini are all LLMs. "Large" refers to the number of parameters (often billions) in the model, which correlates with capability.
LoRA (Low-Rank Adaptation)
A technique for efficiently fine-tuning AI models by training only a small set of additional parameters rather than the entire model. LoRAs are popular in the Stable Diffusion community for creating custom styles and subjects without requiring massive compute resources.
Multimodal AI
AI systems that can process and generate multiple types of content — text, images, audio, video — within a single model. GPT-4o is multimodal (text, image, voice), while earlier models were text-only. Multimodal capability is becoming the standard for new AI tools.
Parameters
The internal variables that an AI model learns during training. The number of parameters (often in the billions for modern LLMs) is a rough indicator of model capability, though architecture and training data quality matter as much or more.
Prompt Engineering
The practice of crafting inputs to AI models to achieve optimal outputs. Effective prompt engineering includes providing context, specifying format, giving examples, and iterating based on results. It is a skill that improves with practice and experimentation.
RAG (Retrieval-Augmented Generation)
A technique that combines an AI language model with an external knowledge source. Instead of relying solely on the model's training data, RAG retrieves relevant documents from a database and feeds them to the model, improving accuracy and reducing hallucinations.
Token
The basic unit of text that AI models process. A token is roughly 4 characters or 0.75 words in English. "Hello world" is two tokens. Token counts determine pricing for API usage and limits for context windows.
Training Data
The dataset used to teach an AI model. The quality, quantity, and diversity of training data directly affect model performance. Concerns about biased training data, copyrighted material in training sets, and data provenance are central to AI ethics and regulation.
Transformer
The neural network architecture that underlies modern language models like GPT, Claude, and Gemini. Introduced in 2017, transformers use attention mechanisms to process relationships between all parts of the input simultaneously, enabling much better language understanding than previous architectures.
Zero-Shot Learning
When an AI model performs a task it was not explicitly trained for, relying on its general knowledge. "Classify this email as spam or not spam" without any training examples is a zero-shot task. Modern LLMs are remarkably capable at zero-shot tasks, though providing examples (few-shot) typically improves results.
Few-Shot Learning
Providing an AI model with a small number of examples (typically 1-5) before asking it to perform a task. "Here are three examples of product descriptions in our brand voice. Now write one for this product." Few-shot prompting significantly improves output quality compared to zero-shot for many tasks.
Keep Learning
Want to go deeper? Check out our Beginner's Guide to AI Tools for a practical introduction to using AI, or browse our tool categories to find AI tools for your specific needs.