Glossary
AI in plain English.
Definitions for the terms you'll see across our prompts and guides.
C
- Context window
- The token limit for information a model can work with in a request. Check the model documentation for how input, output and any internal reasoning tokens count towards its limits.Source
- Chain of thought
- Intermediate reasoning steps used to work towards an answer. Prompting for written steps is model-dependent: OpenAI advises that its reasoning models do not need instructions to think step by step.Source
- Code-switching
- Mixing languages within a sentence or paragraph, common in Singapore (Singlish), Malaysia (Manglish), Philippines (Taglish), and many other multilingual contexts.
E
- Embedding
- A numerical vector representation of content. Text embeddings can support semantic search by comparing representations of a query and candidate passages.Source
F
- Few-shot prompting
- Including a small number of input/output examples in the prompt to teach the model the pattern you want.
- Fine-tuning
- Additional training that adapts a pretrained model for a task or desired behaviour. Whether it helps should be assessed with representative evaluations alongside prompting and retrieval options.Source
H
- Hallucination
- A plausible-sounding but inaccurate or fabricated model output. Providing sources can help, but a citation or confident answer does not establish that a claim is correct.Source
- Honorifics (Korean)
- Korean speech levels (-습니다, -해요, -해, etc.) that mark formality and the speaker-listener relationship. Vital in business communication.
K
- Keigo (敬語)
- Japanese respectful and polite language. Japan's Agency for Cultural Affairs distinguishes five categories: respectful language, humble language I, humble language II, polite language and beautifying language.Source
P
- Prompt
- The instruction or question you send to an AI model. Better prompts produce better outputs; the craft of writing them is sometimes called prompt engineering.
R
- RAG (Retrieval-Augmented Generation)
- An approach that retrieves relevant information and supplies it as context for generation. The retrieved material gives the model evidence to use; it does not by itself verify the answer.Source
S
- System prompt
- An instruction that frames the AI's role, persona, or rules across the whole conversation, separately from the user's individual messages.
T
- Token
- A unit a model uses to process text, such as a character, part of a word, a word or punctuation. The count depends on the model, encoding and language.Source
- Temperature
- A sampling setting, where supported, that changes output variability. Higher values generally increase randomness; lower values favour more predictable choices. It is not a factual-accuracy score.Source