AI Glossary
Plain-English definitions of the AI terms you actually run into, in one place — no jargon left unexplained.
AI Agent
A system that uses an AI model to autonomously plan and take multi-step actions toward a goal — searching, calling tools, or writing code — rather than just answering a single prompt.
API (Application Programming Interface)
A way for one piece of software to call another programmatically. Most AI tools expose an API so developers can use the model directly in their own app, not just through a chat interface.
AlverHub Public API →Chatbot
A conversational AI interface you interact with in natural language — the most common way people first use AI tools.
Chatbots category →Context Window
The maximum amount of text (measured in tokens) a model can 'see' at once, including your prompt, any attached documents, and its own prior replies in the conversation.
Diffusion Model
A type of generative model, commonly used for image generation, that learns to produce an image by starting from random noise and gradually refining it.
Image Generation category →Embedding
A numeric representation of text (or an image) that captures its meaning as a vector, so a computer can compare how similar two pieces of content are — the basis of most AI-powered search and "similar items" features.
Few-Shot Prompting
Giving a model a few examples of the input/output pattern you want directly in the prompt, so it can follow the pattern without any retraining.
Fine-Tuning
Further training an already-trained model on a smaller, specific dataset so it performs better on a narrower task or adopts a particular style — distinct from prompting, which doesn't change the model itself.
Foundation Model
A large, general-purpose model (like GPT, Claude, or Gemini) trained on broad data, that other applications and fine-tuned models are built on top of.
GAN (Generative Adversarial Network)
An older generative-AI architecture where two neural networks — a generator and a discriminator — compete during training, historically used for image generation before diffusion models became more common.
Generative AI
AI systems that create new content — text, images, audio, video, or code — rather than just classifying or predicting from existing data.
GPT (Generative Pre-trained Transformer)
OpenAI's family of large language models, and the technology underlying ChatGPT. Often used loosely to refer to LLM-based chatbots in general, though GPT specifically refers to OpenAI's models.
Hallucination
When an AI model states something false or fabricated with the same confident tone as a correct answer. A core reason to verify important AI output rather than trust it outright, especially for facts, citations, or numbers.
Inference
Running a trained model to produce an output — as opposed to training, which is the (much more expensive) process of building the model in the first place. Every time you send a prompt, you're triggering inference.
LLM (Large Language Model)
A neural network trained on massive amounts of text to predict and generate language, powering most modern AI chatbots and writing tools.
Model
The trained system itself — the file/weights that turn an input into an output. "GPT-4," "Claude," and "Gemini" are all examples of specific models.
Multimodal
A model that can work with more than one type of input or output — for example, understanding both text and images, or generating both text and audio — rather than being limited to text alone.
NLP (Natural Language Processing)
The broader field of AI concerned with understanding and generating human language — the umbrella field that LLMs are the current dominant approach within.
Open-Weight Model
A model whose trained parameters are published for anyone to download and run themselves, as opposed to a closed model only accessible through a company's hosted API.
Open-source AI tools →Parameters
The internal numeric values a model learns during training. Larger parameter counts generally (not always) correlate with more capability, but aren't a complete measure of quality on their own.
Prompt
The input text you give an AI model — your question, instruction, or request. How you phrase it has a real, measurable effect on the quality of the response.
How to Write Better AI Prompts →Prompt Engineering
The practice of deliberately crafting prompts — with context, examples, and structure — to reliably get better output from a model, without changing the model itself.
RAG (Retrieval-Augmented Generation)
A technique where a model looks up relevant information from an external source (a document set, a database, the live web) before answering, so its response can be grounded in real, current data instead of only what it memorized during training.
Reasoning Model
A model variant tuned to work through multi-step problems more deliberately (visible or hidden intermediate reasoning steps) before producing a final answer — typically slower but more accurate on complex logic, math, or coding tasks than a standard chat model.
RLHF (Reinforcement Learning from Human Feedback)
A training technique where human evaluators rate a model's outputs, and that feedback is used to steer the model toward more helpful, accurate, or safe responses.
Temperature
A setting that controls how random or deterministic a model's output is. Lower temperature gives more predictable, focused answers; higher temperature gives more varied, creative ones.
Token
A chunk of text (often a word or part of a word) that a model processes as its basic unit. Both a model's context window and most API pricing are measured in tokens, not words or characters.
Training Data
The (typically enormous) dataset a model learns from before it's released. A model's knowledge, capabilities, and biases are all shaped by what was — and wasn't — in its training data.
Transformer
The neural network architecture (introduced in 2017) underlying essentially every modern LLM. Its key innovation, 'attention,' lets the model weigh how relevant every other word in the input is to each word it's processing.
Zero-Shot Prompting
Asking a model to perform a task with no examples given in the prompt at all — relying entirely on what it already learned during training.