AI Glossary
40 AI terms explained in plain English. No jargon, no PhD required. Click any term for the full definition with practical context.
A
AGI
The idea of a much more general AI that can do almost everything a capable person can do intellectually.
AI Agent
An AI that can actually do things for you, not just talk about them.
API
The behind-the-scenes way apps plug AI into their own product.
Attention
How the model decides what parts of the text to pay attention to right now.
Attention Head
One of the model's separate mini-focus channels inside an attention layer.
B
C
Chain-of-Thought
Asking the model to show its working instead of jumping straight to an answer.
Context Rot
When a long AI session gets messy and starts carrying around too much old baggage.
Context Window
The model's short-term memory size for one conversation turn.
F
Feed-Forward Network (FFN)
The part of each layer that reshapes what the model has learned after it looks across the sequence.
Fine-tuning
Teaching a strong general model to become better at your specific job.
Function Calling
How a model asks software to actually do something, like search, fetch, or send.
H
L
M
O
P
Prompt
The message you send to the AI.
Prompt Caching
If you keep sending the same starting prompt, the provider may reuse it instead of reprocessing everything from scratch.
Prompt Engineering
Learning how to ask AI the right way so it gives better results.
R
RAG
Letting the AI look things up before it answers.
Reasoning Model
A model built to think longer and more carefully on difficult questions.
Residual Connection
A shortcut that lets a layer improve the signal without throwing away what came before.
RLHF
Training the model based on which answers people preferred.
T
Temperature
A creativity dial for the model's responses.
Token
The small pieces of text AI counts, reads, and bills you for.
Tokenization
How your text gets chopped into the small pieces the model really sees.
Tokenizer
The thing that decides how your prompt gets broken into billable pieces.
Tool Use
When the AI can do more than talk and can actually reach out to software to get work done.
Transformer
The core invention that made today's powerful AI models possible.
Transformer Block
One repeatable layer inside a modern model stack.
TTFT (Time to First Token)
How long you wait before the model starts answering at all.