AI terms, in everyday language
A short glossary for practical business conversations.
LLM
A language model trained to generate and transform text. It predicts plausible output; it does not guarantee truth.
RAG
Retrieval-augmented generation: search approved information first, then use relevant passages to draft an answer. Source access and quality still matter.
Fine-tuning
Further training a model on examples to change behaviour. It is not the default way to give a model current company knowledge.
Agent
A workflow in which a model can choose actions or tools. More autonomy needs tighter permissions, evaluation and recovery controls.
Hallucination
An unsupported or incorrect output that may sound convincing. Verify against evidence.
Tokens
Small chunks used to process text. Providers may meter input and output tokens; task cost also includes review and infrastructure.
Embeddings
Numerical representations used to find similar text. Similarity does not prove factual relevance.
Prompt injection
Instructions hidden in untrusted input that try to redirect a model or workflow. Treat external content as data, not authority.
Assess your starting point yourself, or work through it together.
Take the free assessment →