GLOSSARY

Generative AI

Generative AI produces new content — text, image, code, audio — using transformer or diffusion models, adapted via prompting, RAG, fine-tuning, or agents.

Last updated:

Quick answer
Generative AI refers to a class of machine learning models that produce new content — text, images, audio, video, code, or structured data — rather than classify or predict against fixed outputs. The current dominant architectures are transformer-based large language models and diffusion models, deployed in enterprise settings via APIs, private cloud, or self-hosted open-weight models depending on compliance needs.

WHAT IT IS

Generative AI systems are trained on large corpora and can be adapted to specific tasks through prompting, in-context examples, retrieval augmentation (RAG), fine-tuning, or agentic tool use. Production systems layer these with evaluation harnesses, content safety filters, and human-in-the-loop review.

HOW IT WORKS

Enterprise deployments cluster around six patterns: internal assistants (policy, HR, IT), knowledge retrieval over proprietary content, customer-facing chat with guardrails, content generation for marketing, code generation for engineering, and process automation where an agent chains tool calls. Quality, cost, latency, and risk trade off differently at each pattern.

WHEN TO USE

Deploy generative AI when work involves pattern-rich natural language or media, tolerates probabilistic outputs, or benefits from iteration. Avoid it where deterministic accuracy is required and no human review exists.

RELATED

SOURCES

Related questions.

What is generative AI?
Generative AI is a class of machine-learning models that produce new content — text, images, audio, video, code, or structured data — rather than classify or predict against fixed outputs. The current dominant architectures are transformer-based large language models for text and code, and diffusion models for images and video.
How does generative AI work?
Models are pre-trained on massive datasets to learn the statistical structure of the content they will generate. At inference time, they produce new outputs by sampling from that learned distribution, conditioned on the user's prompt. Fine-tuning and retrieval augmentation steer the model toward specific tasks or corpora.
What can generative AI be used for in the enterprise?
Content drafting, summarization, translation, customer-service triage, code generation and review, knowledge retrieval over internal documents, image and video creation, and structured-data extraction from unstructured sources. High-value use cases share a common shape: bounded domain, clear evaluation criteria, and a reviewer in the loop.
What are the main risks of generative AI?
Hallucination (confident incorrect output), prompt injection, data leakage via model context, copyright exposure in outputs, and model drift as providers change behavior. Production deployments need evaluation sets, retrieval grounding, access controls, and monitoring — not just a well-written prompt.
How does NUUN AI build generative AI applications?
We scope builds to a named use case, ground the model in retrieval over client data, evaluate against a held-out test set before shipping, and monitor quality in production. We align every production deployment to NIST AI RMF and ISO/IEC 42001.

Need this term in action?

Book a call