AI Jargon Buster
Confused by AI terms in the news? Search any term and get a simple, plain-English explanation. Free AI dictionary.
Artificial Intelligence
Software that can perform tasks which normally need human intelligence — like understanding language, recognizing images, or making decisions.
Machine Learning
A type of AI where the software learns patterns from data instead of being hand-programmed with rules.
Deep Learning
Machine learning using large neural networks with many layers — the approach behind today’s most powerful AI.
Neural Network
A system of connected ‘neurons’ (math units) that learns from examples, loosely inspired by the brain.
Algorithm
A step-by-step set of instructions a computer follows to solve a problem.
Dataset
A collection of data (text, images, etc.) used to train or test an AI model.
Training
The process of feeding a model lots of data so it learns patterns — like studying before an exam.
Inference
When a trained model is actually used to answer questions or make predictions.
Parameter
A tunable value inside a model learned during training. More parameters usually means a more capable (and expensive) model.
Token
The chunk of text a model reads or writes at a time — roughly ¾ of an English word. ‘ChatGPT’ is 2 tokens.
Prompt
The instruction or question you give an AI to get it to do something.
Prompt Engineering
The skill of writing good prompts to get better, more reliable results from AI.
Context Window
How much text a model can ‘remember’ at once. A bigger window means longer conversations and documents.
Hallucination
When an AI confidently makes up facts that aren’t true. Always verify important claims.
Bias
When a model’s outputs unfairly favor or disfavor certain groups — usually inherited from its training data.
Large Language Model (LLM)
An AI trained on huge amounts of text that can understand and generate human-like language. ChatGPT and Claude are LLMs.
Transformer
The neural-network design (from 2017) behind nearly all modern LLMs.
GPT
‘Generative Pre-trained Transformer’ — OpenAI’s family of language models (GPT-4, GPT-5, …).
Diffusion Model
The technique behind AI image generators: start from noise and gradually refine it into a picture.
Fine-tuning
Taking a pre-trained model and training it a bit more on specific data to specialize it.
Pre-training
The first, massive training phase where a model learns general knowledge from huge datasets.
Foundation Model
A large general-purpose model (like GPT-4 or Claude) that other apps and tools are built on top of.
Open-weights Model
A model whose internal parameters are publicly downloadable (e.g. Llama, Mistral) — you can run and modify it yourself.
Multimodal
A model that handles more than one type of input — text plus images, audio, or video.
Embedding
A way of turning words or images into lists of numbers so the AI can measure how similar things are.
Vector Database
A database optimized for storing embeddings, used to find semantically similar content fast.
Reinforcement Learning
Training where the AI learns by trial and error, getting rewards for good actions — like training a pet.
RLHF
‘Reinforcement Learning from Human Feedback’ — humans rate the model’s answers so it learns to be more helpful and safe.
RAG
‘Retrieval-Augmented Generation’ — the AI looks up relevant documents before answering, so answers are fresher and more accurate.
Chain-of-Thought
Asking the model to reason step-by-step, which improves answers to hard problems.
Few-shot Learning
Giving the model a few examples in the prompt so it copies the pattern.
Zero-shot Learning
The model handles a task with no examples at all — just the instruction.
Temperature
A setting controlling randomness: low = focused and predictable, high = creative and surprising.
System Prompt
Hidden instructions set by the developer that shape how the AI behaves in every reply.
AI Agent
An AI that doesn’t just answer — it takes actions: browsing, using tools, and completing multi-step tasks on its own.
Tool Use / Function Calling
Letting a model call external tools (search, calculators, apps) to get things done beyond chatting.
Quantization
Shrinking a model so it runs on smaller devices, with only a small quality trade-off.
Streaming
When the AI’s answer appears word-by-word as it’s generated, instead of all at once.
Alignment
Making sure AI systems pursue goals that match human values and intentions.
Red Teaming
Deliberately trying to break or trick an AI to find weaknesses before bad actors do.
Guardrails
Built-in limits that stop a model from producing harmful or unwanted outputs.
Jailbreak
A trick prompt that tries to bypass a model’s safety guardrails.
Deepfake
AI-generated fake images, audio, or video that look like a real person.
Synthetic Data
Artificially generated training data — useful, but overuse can degrade models.
AI Safety
The field studying how to keep advanced AI systems safe, controllable, and beneficial.
Copyright (AI Training)
The ongoing legal debate over whether training AI on copyrighted text and images is fair use.
AGI
‘Artificial General Intelligence’ — a hypothetical AI that can do any intellectual task a human can. It doesn’t exist yet.
Compute
The processing power (usually GPUs) needed to train and run AI models — one of AI’s biggest costs.
API
A way for apps to use an AI model programmatically — how most AI features inside other apps work.
Latency
The delay between asking the AI something and getting a response.
Benchmark
A standardized test used to compare how capable different AI models are.
OpenAI
The company behind ChatGPT and GPT models, led by Sam Altman.
Anthropic
The AI safety-focused company behind the Claude models, led by Dario Amodei.
xAI
Elon Musk’s AI company, maker of the Grok models.
Google DeepMind
Google’s AI lab behind the Gemini models and AlphaGo.
Edge AI
Running AI directly on your device (phone, laptop) instead of in the cloud — faster and more private.