Clear definitions of 49 AI terms — LLM, GPT, RAG, prompt engineering, fine-tuning, and more. Written for beginners, updated for 2026.
Core AI Concepts
These are the foundational terms you'll see across nearly every AI product page and review — the vocabulary that everything else builds on.
AGI (Artificial General Intelligence)
A hypothetical type of AI that can understand, learn, and apply knowledge across any intellectual task a human can perform.
AI Agent
An autonomous AI system that can perceive its environment, make decisions, and take actions to accomplish specific goals.
Deep Learning
A subset of machine learning that uses neural networks with many layers to learn complex patterns from large amounts of data.
NLP (Natural Language Processing)
The field of AI focused on enabling computers to understand, interpret, and generate human language.
Multimodal AI
AI systems that can process and generate multiple types of data — such as text, images, audio, and video — within a single model.
Zero-Shot Learning
A technique where an AI model performs a task it was never explicitly trained on, using only a natural language instruction.
Language Model Architecture
Modern AI tools are built on the Transformer architecture, introduced in the 2017 paper "Attention Is All You Need". These terms describe how large language models are structured and how they represent meaning internally.
The neural network architecture behind virtually all modern language models, introduced in the 2017 paper "Attention Is All You Need."
LLM (Large Language Model)
A neural network trained on massive amounts of text data that can understand, generate, and reason about human language.
Foundation Model
A large AI model trained on broad, diverse data that can be adapted to a wide range of downstream tasks. GPT-4, Claude, Llama, and Gemini are all foundation models.
A family of large language models developed by OpenAI that generate text by predicting the next token in a sequence.
Embedding
A numerical representation of text, images, or other data as a dense vector in a high-dimensional space. Embeddings capture semantic meaning.
Vector Database
A specialized database optimized for storing and searching high-dimensional vector embeddings.
Training & Fine-Tuning
A base model gets adapted for specific tasks through the techniques below. According to OpenAI's own research on instruction-following, human-feedback training is what makes a raw language model behave like a helpful assistant rather than just predicting the next word.
Fine-Tuning
The process of taking a pre-trained AI model and further training it on a smaller, task-specific dataset to improve its performance.
LoRA (Low-Rank Adaptation)
A parameter-efficient fine-tuning technique that adds small, trainable adapter layers to a frozen pre-trained model instead of updating all its weights.
RLHF (Reinforcement Learning from Human Feedback)
A training technique where human evaluators rank AI outputs by quality, and those rankings are used to train a reward model.
Chain-of-Thought (CoT)
A prompting technique that encourages a language model to break down complex reasoning into intermediate steps before arriving at a final answer.
Retrieval, APIs & Tooling
These terms cover how AI tools connect to outside data and to each other — the plumbing behind features like "search the web" or "connect your documents."
RAG (Retrieval-Augmented Generation)
A technique that enhances AI responses by first retrieving relevant documents from an external knowledge base, then feeding that context to the language model.
API (Application Programming Interface)
A set of rules and protocols that allows different software applications to communicate with each other.
Token
The basic unit of text that language models process — typically a word, part of a word, or punctuation mark. LLM pricing, context limits, and speed are all measured in tokens.
Common AI Pitfalls & Techniques
Understanding these terms helps you evaluate AI tool output critically, rather than trusting it by default.
Hallucination
When an AI model generates information that sounds plausible but is factually incorrect or entirely fabricated.
Diffusion Model
A type of generative AI model that creates images by learning to reverse a gradual noising process. DALL·E, Midjourney, and Stable Diffusion all use this approach.
Prompt Engineering
The practice of crafting effective instructions (prompts) to get the best possible output from an AI model.
What does "hallucination" actually look like in practice?
Answer: a hallucination is when an AI model states something false with the same confidence as something true, and the fix is always to verify against a primary source, not to trust fluent-sounding output.
Context: hallucinations happen because a language model is predicting statistically likely text, not looking up facts, unless it's specifically using retrieval-augmented generation (RAG) to ground its answer in real documents.
Example: asking a model "what year was AI Nexus founded?" without giving it that information risks a confident but invented answer, since the model has no ground truth to retrieve — the same way it would invent a plausible-sounding citation for a fact it was never trained on.
Frequently asked AI terminology questions
What is AGI?
A hypothetical type of AI that can understand, learn, and apply knowledge across any intellectual task a human can perform.
What is AI Agent?
An autonomous AI system that can perceive its environment, make decisions, and take actions to accomplish specific goals.
What is API?
A set of rules and protocols that allows different software applications to communicate with each other.
What is Chain-of-Thought?
A prompting technique that encourages a language model to break down complex reasoning into intermediate steps before arriving at a final answer.
What is Deep Learning?
A subset of machine learning that uses neural networks with many layers to learn complex patterns from large amounts of data.
What is Diffusion Model?
A type of generative AI model that creates images by learning to reverse a gradual noising process. DALL·E, Midjourney, and Stable Diffusion all use this approach.
What is Embedding?
A numerical representation of text, images, or other data as a dense vector in a high-dimensional space. Embeddings capture semantic meaning.
What is Fine-Tuning?
The process of taking a pre-trained AI model and further training it on a smaller, task-specific dataset to improve its performance.
What is Foundation Model?
A large AI model trained on broad, diverse data that can be adapted to a wide range of downstream tasks. GPT-4, Claude, Llama, and Gemini are all foundation models.
What is GPT?
A family of large language models developed by OpenAI that generate text by predicting the next token in a sequence.