Generative AI

Generative AI refers to models that create new content—text, images, audio, video, or code—conditioned on prompts or other inputs.

These systems produce new material instead of only classifying or scoring what already exists.

Generative models learn distributions over content and sample new instances. Approaches include autoregressive Transformers for text/code and diffusion models for images, among others.

Generative products changed enterprise software roadmaps after 2022–2023. Buyers must distinguish generation from classical classification or rules.

• Drafting and rewriting content
• Image and media generation
• Synthetic data for testing (with caveats)

• Accelerates content and design iteration
• Enables new UX patterns (chat, copilots)
• Can bootstrap data when carefully validated

• IP, privacy, and brand-risk concerns
• Quality variance and hallucinations
• Evaluation is harder than for fixed classifiers

• Generative does not mean unsupervised or ungoverned
• Pretty outputs can still be wrong

Does Generative AI always mean ChatGPT-style chat?

No. Generation includes images, code, audio, and video. Chat is one product pattern.

• NIST AI RMF

• Goodfellow et al. — Generative Adversarial Networks (historical generative method)