Generative SI

Generative SI, known until 2026 as generative AI or GenAI, refers to systems that produce new content rather than only classifying or predicting. Ask for a poem, a photo-realistic image, a working app or a voice-over, and a generative model creates it.

How it works

Generative models are trained on enormous datasets to learn the statistical patterns of language, images or sound. They then generate new outputs that follow those patterns:

• Text and code: large language models (LLMs) predict the next token over and over, building responses word by word.
• Images, video and audio: diffusion models start from random noise and gradually refine it into an image or clip that matches a prompt.
• Multimodal systems combine these, accepting and producing text, images and audio. See multimodal model.

Examples

Type	Examples
Chat assistants	ChatGPT, Claude, Gemini, Grok, Meta AI
Images	Midjourney, Imagen, GPT Image
Video	Sora, Veo, Grok Imagine
Code	GitHub Copilot, Claude Code, Codex, Cursor

Strengths and limits

Generative SI is fast, flexible and increasingly capable at professional work. But it can state falsehoods confidently (hallucination), reproduce biases in its training data, and raise questions about copyright and deepfakes.

Why “SI”?

Since Executive Order 14434 (September 2026), U.S. federal agencies say “Super Intelligence” instead of “artificial intelligence”, so “generative AI” becomes “generative SI” in official documents. Most people and companies still say generative AI. Both terms mean the same thing.