📖 AI Reference Guide

Artificial Intelligence Glossary

Clear, jargon-free definitions for the most important concepts, technical models, and frameworks shaping modern artificial intelligence in 2026.

AI Agent
An autonomous software entity powered by an LLM that can perceive its environment, formulate multi-step plans, execute actions using tools/APIs, and self-correct until a goal is achieved.
Large Language Model (LLM)
A deep learning model trained on massive multilingual text corpora that can generate, analyze, translate, summarize, and reason through natural human language.
Retrieval-Augmented Generation (RAG)
An architecture that retrieves relevant documents from an external vector database before passing them to an LLM, ensuring factual accuracy and eliminating hallucinations on private data.
Diffusion & Flow Matching Models
Generative architectures that synthesize high-fidelity images, audio, or video by iteratively reversing a noise process — the core engine behind Midjourney, Flux.1, and Stable Diffusion.
Context Window
The maximum number of tokens (words and characters) that an AI model can retain and analyze in active memory during a single prompt interaction.
Vector Embedding
A high-dimensional mathematical coordinate vector representing the semantic meaning of text, audio, or images, enabling lightning-fast similarity search.
Fine-Tuning (SFT & LoRA)
The process of taking a pre-trained base model and training it further on a specialized domain dataset to adopt a specific tone, task, or organizational knowledge base.
Quantization (GGUF / AWQ)
A compression technique that reduces the precision of model weights (e.g. from 16-bit to 4-bit) so powerful models can run on consumer laptops and GPUs with minimal loss of intelligence.
Multimodal Model
An AI model capable of seamlessly understanding, processing, and generating multiple data modalities simultaneously, such as text, images, video, and spoken audio.
Prompt Engineering
The practice of structuring, phrasing, and constraining natural language inputs, system instructions, and few-shot examples to achieve optimal responses from an AI model.
Hallucination
When a generative AI model outputs factually incorrect statements, fabricated citations, or non-existent code libraries with absolute confidence.
Tool Use / Function Calling
The capability of an LLM to emit structured parameters (like JSON) to trigger external APIs, database queries, calculators, or web searches to accomplish real-world tasks.
Inpainting & Outpainting
Image editing techniques where AI replaces or modifies a specific masked area of an image (inpainting) or seamlessly extends the canvas beyond its original borders (outpainting).
Open Weights Model
An AI model whose neural network parameters and weights are made publicly downloadable, allowing developers to inspect, fine-tune, and host the software locally on their own hardware.
Voice Cloning (TTS)
Using deep neural audio models to analyze vocal timbre, pitch inflection, and cadence from a short audio sample to generate synthetic speech in the exact voice of the speaker.
Temperature (Sampling)
A hyperparameter controlling randomness in generative models. A lower temperature (e.g. 0.2) produces deterministic, factual outputs, while a higher temperature (e.g. 0.8) yields creative variations.

Want to put these concepts into practice?

Explore thousands of verified tools utilizing these cutting-edge architectures across 100+ categories.

Join 50,000+ AI readers. Stay ahead of the latest tools.

Get one useful email each week with new AI tools, launches, and practical insights — no fluff, just what's worth knowing. It's free, low-frequency, and easy to unsubscribe anytime.

AS READ BY PEOPLE AT

Microsoft Google Amazon Meta Intel