Glossary — Generative AI: what can it do for your business?
Key terms explained
Explore this guide : FAQ Checklist Case study Tools & resources Definition Benefits Complete guide Use cases For businesses Statistics & data Comparison Cost & budget Training & learning Trends Common mistakes Alternatives How to write about it Read the full guide
- LLM (Large Language Model)
- A neural network trained on large text corpora to generate, classify, or transform text. Examples include GPT-4, Claude, and Gemini. LLMs predict the most probable next token given a context window; they do not retrieve live information unless connected to external tools.
- Prompt
- The instruction or context you send to a language model to guide its response. Prompt quality has a large impact on output quality: include the role, task, constraints, and expected output format. A system prompt sets default behaviour; a user prompt contains the specific request.
- RAG (Retrieval-Augmented Generation)
- An architecture that searches a knowledge base for relevant passages before passing them as context to a language model. RAG grounds answers in your actual documents without retraining the model, and allows you to cite sources.
- Token
- The basic unit of text a language model processes — roughly a word or word fragment. Models have a context window measured in tokens. Input and output tokens are both billed by API providers.
- Hallucination
- When a language model generates plausible-sounding but factually incorrect information. It occurs because LLMs optimise for textual coherence, not factual accuracy. Mitigate by providing grounding context, validating outputs against authoritative sources, and keeping humans in the loop.
- Fine-tuning
- Continuing the training of a pre-trained model on a curated dataset to adjust its behaviour for a specific task or style. More expensive and slower to iterate than prompting. Justified only when the desired behaviour cannot be achieved with prompt engineering and RAG alone.
TUPLIFY STUDIO
Ready to build?
Tuplify Studio turns these concepts into working software.
Discuss your project