Core Terms
Token
The basic unit of text a model processes — roughly a word or part of a word. Pricing and context limits are usually measured in tokens.
Context Window
The maximum amount of text (in tokens) a model can consider at once in a single conversation or request.
Parameters
The internal numerical values a model learns during training; loosely correlated with capability but not a precise capability measure on its own.
Fine-Tuning
Additional training on a specific dataset to specialize a pre-trained model for a narrower task or domain.
Hallucination
When a model generates confident-sounding but factually incorrect or fabricated information.
Inference
The process of running a trained model to generate a response to a real query, as opposed to training it.
Temperature
A setting controlling how random or deterministic a model's output is; lower values produce more predictable, focused responses.
System Prompt
Instructions given to a model before a conversation starts, shaping its behavior and tone for the rest of the interaction.
Architecture & Technique Terms
Transformer
The neural network architecture underlying most modern language models, built around a mechanism called attention.
Mixture of Experts (MoE)
An architecture where only a subset of a model's total parameters activate for any given input, improving efficiency at large scale.
Embedding
A numerical representation of text, image, or other content that captures its meaning for similarity search; see our Vector Databases page.
RAG
Retrieval-Augmented Generation — letting a model pull in external information before responding; see our RAG page.
Where Else to Look
This glossary covers the most common terms; for deeper explainers on specific concepts, see our dedicated pages on MCP, RAG, vector databases, and prompt engineering.
Related Pages
Frequently Asked
What's the difference between a token and a word?
A token is roughly a word or part of a word — common words are often a single token, while longer or less common words may be split into multiple tokens.
Why does 'hallucination' matter?
It's one of the most important limitations to understand when using any AI model — confident-sounding output isn't automatically accurate, so verifying anything factually important is good practice.
What does 'parameters' actually tell you about a model?
It's a rough size indicator, loosely correlated with capability, but not a precise measure — a smaller, well-trained model can outperform a larger, poorly-trained one on specific tasks.
Is this glossary exhaustive?
It covers the terms you'll encounter most often on this site; for a specific term not listed here, our dedicated concept pages (linked throughout) go deeper on individual topics.