← Reference · Nestor G Pestelos Jr · Print this page

Artificial Intelligence · Generative Models

Generative AI

Reference entry · last updated 20260908

Generative artificial intelligence (generative AI or GenAI) comprises models that produce synthetic content based on patterns in data. Outputs can include text, images, audio, video, and code.[1]

First principles and definitions

A generative model represents patterns that allow it to produce content. Training adjusts the model using data; inference uses the trained model to generate an output. In a text model, a prompt supplies context for generation.

Use in software development

Large language models can generate source text. A coding agent combines a model with tools and an execution loop, allowing it to inspect files, propose changes, and run checks. Generating text and executing a change are separate operations.

Verification limits

Generated code needs review and testing against requirements. A plausible output can contain errors or security weaknesses. NIST’s secure development profile addresses practices for developing generative AI and foundation models; it is not evidence that any generated program is correct.[2]

See also

• Acceleration Whiplash
• Artificial Intelligence
• Large Language Models
• AI Agents
• LLM Evaluations

References

1. NIST CSRC Glossary. Generative artificial intelligence.
2. NIST, 2024, SP 800-218A. Secure Software Development Practices for Generative AI and Dual-Use Foundation Models.