AI jargon can feel like a foreign language. You read job postings mentioning "RAG architecture," LinkedIn posts about "transformer attention heads," and course descriptions covering "RLHF." If you do not know what these terms mean, you feel excluded from the conversation.
This glossary defines every important generative AI term in simple, beginner-friendly language. Bookmark this page and come back whenever you encounter a term you do not recognise.
A-D
Agentic AI
AI systems that can autonomously plan, reason, use tools, and complete multi-step tasks without constant human guidance. Unlike chatbots that respond to individual prompts, AI agents break down goals and execute complex workflows independently.
AI Hallucination
When an AI model generates information that sounds plausible but is factually incorrect. This happens because LLMs predict likely text rather than retrieving verified facts. RAG systems help reduce hallucinations by grounding answers in actual documents.
Attention Mechanism
The core innovation inside transformer models that allows AI to determine which words in a text are most relevant to each other. It is what enables LLMs to understand context and relationships between words, regardless of how far apart they are.
Chain-of-Thought (CoT)
A prompting technique where you ask the AI to reason step by step before giving a final answer. This significantly improves accuracy on complex tasks like math, logic, and multi-step analysis.
Chunking
The process of splitting large documents into smaller pieces (chunks) for storage in a vector database. Chunk size affects RAG system performance: too large and retrieval becomes imprecise, too small and context is lost.
Context Window
The maximum amount of text an LLM can process at once, including both your input and its output. Think of it as the model's working memory. Larger context windows allow processing longer documents and conversations.
Diffusion Model
A type of generative AI model used primarily for image generation. It works by learning to remove noise from images, allowing it to generate new images from random noise. DALL-E, Midjourney, and Stable Diffusion all use diffusion models.
E-H
Embedding
A numerical representation of text (or images) as a list of numbers (vector) that captures its meaning. Words with similar meanings have similar embeddings. Embeddings are the foundation of semantic search and RAG systems.
Few-Shot Prompting
A prompting technique where you provide 2-3 examples of the input-output pattern you want before asking the AI to handle a new input. This helps the model understand exactly what you expect.
Fine-Tuning
The process of further training a pre-trained model on your own data to customise its behaviour for a specific domain or task. Techniques like LoRA and QLoRA make fine-tuning practical with limited computing resources.
Foundation Model
A large AI model trained on broad data that can be adapted for many different tasks. GPT-4, Claude, and Llama are all foundation models. They serve as the "foundation" upon which specific applications are built.
Generative AI
A type of artificial intelligence that creates new content (text, images, code, audio, video) by learning patterns from existing data. It is different from traditional AI, which classifies or predicts rather than creates.
GPU (Graphics Processing Unit)
Specialised hardware originally designed for rendering graphics that is now essential for training and running AI models. NVIDIA GPUs are the industry standard for AI workloads.
I-L
Inference
The process of using a trained model to generate outputs. When you send a prompt to ChatGPT and receive a response, that is inference. Training teaches the model; inference is the model using what it learned.
LangChain
The most popular open-source framework for building LLM-powered applications. It provides pre-built components for RAG, agents, memory, and tool integration.
Large Language Model (LLM)
An AI model trained on massive amounts of text data to understand and generate human language. "Large" refers to both the training data (billions of words) and the model size (billions of parameters). GPT-4, Claude, and Llama are all LLMs.
LoRA (Low-Rank Adaptation)
A technique for fine-tuning large language models efficiently by training only a small number of additional parameters rather than the entire model. This makes fine-tuning practical on consumer-grade hardware.
M-P
Multimodal
AI models that can process and generate multiple types of content (text, images, audio, video) rather than just one. GPT-4o and Gemini are multimodal models.
Neural Network
A computing system inspired by the human brain, consisting of interconnected nodes (neurons) organised in layers. Neural networks are the foundation of deep learning and all modern AI models.
Parameter
A numerical value inside an AI model that is adjusted during training. More parameters generally means a more capable model. GPT-4 is estimated to have over 1 trillion parameters.
Prompt
The text instruction you give to an AI model. It can be a question, a command, or any text that guides the model's response.
Prompt Engineering
The skill of crafting precise, effective prompts to get the best possible output from AI models. It involves techniques like zero-shot, few-shot, chain-of-thought, and role-based prompting.
Q-T
QLoRA
An even more efficient version of LoRA that uses quantisation (reducing the precision of numbers) to further reduce memory requirements for fine-tuning. It makes fine-tuning possible on GPUs with as little as 16GB of memory.
RAG (Retrieval Augmented Generation)
A technique that enhances AI responses by retrieving relevant documents from an external knowledge base before generating an answer. RAG reduces hallucinations and allows AI to work with current and private data.
RLHF (Reinforcement Learning from Human Feedback)
A training technique where human evaluators rank AI responses, and the model learns to produce outputs that humans prefer. RLHF is what makes models like ChatGPT helpful and conversational rather than just predicting random text.
Semantic Search
Search based on meaning rather than exact keyword matches. When you ask "What is our return policy?" semantic search finds documents about "refund guidelines" even if those exact words were not used.
Temperature
A setting that controls how creative or conservative an AI model's responses are. Low temperature (0.1-0.3) produces focused, deterministic responses. High temperature (0.7-1.0) produces more creative, varied outputs.
Token
The basic unit of text that LLMs process. Tokens can be words, parts of words, or punctuation. One token is roughly 3/4 of a word in English. Context windows and API pricing are measured in tokens.
Transformer
The neural network architecture that powers all modern LLMs. Introduced in 2017, it processes all words in a text simultaneously using attention mechanisms, enabling much better understanding of context and relationships.
U-Z
Vector
A list of numbers that represents the meaning of a piece of text (or image) in mathematical space. Vectors are used in embeddings and vector databases to enable semantic similarity search.
Vector Database
A specialised database designed to store and search through vectors efficiently. Instead of searching by exact matches (like SQL databases), vector databases find data that is semantically similar. Pinecone, ChromaDB, and Weaviate are popular vector databases.
Zero-Shot Prompting
Asking an AI to perform a task without providing any examples. The model relies entirely on its training to understand what you want. Works well for straightforward tasks.
This glossary covers the terms you will encounter most frequently in your generative AI learning journey. As you progress, each of these concepts will become second nature.
Frequently Asked Questions About AI Terminology
What does LLM mean?
LLM stands for Large Language Model. It is an AI system trained on massive amounts of text data to understand, generate, and reason with human language. Examples include GPT-4, Claude, Gemini, and Llama.
What is a token in AI?
A token is the basic unit of text that AI models process. Tokens can be whole words, parts of words, or punctuation marks. In English, one token is roughly three-quarters of a word. API pricing and context windows are measured in tokens.
What is fine-tuning?
Fine-tuning is the process of further training a pre-trained AI model on your own data to customise it for a specific task or domain. For example, fine-tuning a general model on medical literature to create a healthcare-specific AI assistant.
What is hallucination in AI?
AI hallucination occurs when a model generates information that sounds plausible but is factually incorrect. It happens because LLMs predict likely text rather than looking up verified facts. RAG systems address this by grounding responses in actual retrieved documents.
Start Your Career in AI Today
Join learners who transformed their careers with our industry-aligned Gen AI program