THE CORE AIM OF VOCABULARY MASTERY · GENERATIVE AI VOCABULARY · PROMPT → CONTEXT → GENERATE → VERIFY → ITERATE
Generative AI vocabulary is the language used to describe artificial intelligence systems that create text, images, audio, video, code and other content. Terms such as prompt, context window, token, generation, hallucination, grounding, retrieval and multimodal matter because generative systems are most useful when people understand both their capabilities and their limits.
The core aim of vocabulary mastery for generative AI vocabulary is generation-and-verification clarity. Learners should be able to describe what information the model receives, how output is generated, where unsupported claims can appear and how evidence or tools can verify important results.
This page is the Generative AI Vocabulary owner inside the eduKateSG Vocabulary hub, alongside Artificial Intelligence Vocabulary and Machine Learning Vocabulary.
Central proposition: Generative AI vocabulary is mastered when the learner can distinguish the instruction, available context, generated output and evidence used to judge that output.
The Generative AI Vocabulary Router
- Input: prompt, instruction, context, token.
- Generation: completion, sampling, output.
- Knowledge: grounding, retrieval, source, citation.
- Failure: hallucination, omission, inconsistency, bias.
- Capability: multimodal, tool use, structured output.
- Evaluation: factuality, relevance, robustness, human review.
Prompt and Context Are Different
A prompt is the immediate instruction or user input. Context is the broader information available during generation, which may include prior messages, documents, retrieved passages, system instructions or tool results.
Hallucination and Verification
A generative model may produce fluent information that is unsupported or incorrect. This is commonly called a hallucination. Important claims should be checked against reliable evidence rather than accepted because the wording sounds confident.
Grounding and Retrieval
Grounding connects generation to external information such as supplied documents, databases or search results. Retrieval is the act of finding relevant information for the model to use. Grounding can improve factual reliability but does not remove the need for evaluation.
Multimodal Vocabulary
Multimodal systems can work across more than one type of input or output, such as text and images. The term describes modality coverage, not guaranteed competence on every task.
How to Learn Generative AI Vocabulary
- Compare prompts with wider context.
- Ask for evidence and verify important claims.
- Compare grounded and ungrounded tasks.
- Identify failure modes explicitly.
- Separate model capability from model reliability.
Common Generative AI Vocabulary Mistakes
Treating generation as retrieval
Repair: distinguish producing an output from fetching an existing record.
Treating fluent language as evidence
Repair: judge claims by sources and validation, not style.
Assuming grounding guarantees truth
Repair: verify whether the model interpreted and used the source correctly.
Frequently Asked Questions
What is generative AI vocabulary?
It is the specialised language used for prompts, context, generation, grounding, multimodal systems, failure modes and evaluation.
What terms should beginners learn first?
Start with prompt, context, token, generation, hallucination, grounding, retrieval, multimodal and evaluation.
What is grounding?
It is connecting model generation to external information or evidence.
The Generative AI Vocabulary Standard
Generative AI vocabulary reaches its core aim when the learner can explain what the model received, what it generated, what evidence supports the result and what still needs verification.
That is the standard: AI language precise enough to keep generation separate from truth.