DataTerminology.comNEW
AI systems that generate novel content—text, images, audio, or code—by learning patterns from training data and sampling from learned distributions.
DESCRIPTION
Generative AI models learn data distributions to create new content, from text (LLMs) to images (diffusion, GANs). They enable assistants, design tools, and synthetic data generation. Capabilities depend on training data quality, architectures, and prompt design. Risks include hallucinations, bias, and safety concerns, requiring evaluation, guardrails, and governance. Production use integrates RAG, fine-tuning, monitoring, and cost controls. Organizations align with ethical frameworks and emerging regulations.
ALTERNATIVE DEFINITIONS
Microsoft Learn
Generative AI is a type of artificial intelligence that can create new content and ideas, including conversations, stories, images, videos, and music.
Enumerates content modalities and creative capabilities.
Google Cloud
Generative AI is a type of artificial intelligence that can create text, images, audio, and synthetic data.
Highlights multimodal outputs and synthetic data generation.
IBM
Generative AI refers to deep-learning models that can generate high-quality text, images, and other content based on the data they were trained on.
Stresses deep learning basis and quality of generated content.
RELATED TERMS
Large Language Model Diffusion Models GAN Prompt Engineering Fine-Tuning Retrieval-Augmented Generation Hallucination Model Evaluation Safety Alignment Synthetic Data
COMPETING TERMS
Predictive AI
RELATIONS
- Narrower: Audio Generation Boltzmann Machine
KNOWLEDGE AREAS
AI Data Science AI Governance MLOps
TARGET AUDIENCE
AI Engineer Data Scientist MLOps Engineer AI Product Manager Chief Data Officer
EXTERNAL SOURCES
- Wikipedia - Generative artificial intelligence
- Microsoft Learn - Overview and architecture guidance
- Google Cloud - Vertex AI generative overview
- arXiv - GAN original paper
- arXiv - DDPM diffusion models paper
TOP QUESTIONS
What makes Generative AI different from predictive AI?
Generative AI produces novel content by sampling learned distributions, while predictive AI forecasts or classifies outcomes from inputs.
Category: Misunderstanding
How do LLMs generate text?
They model token probabilities and sample next tokens conditioned on context, guided by temperature, top-k, or nucleus sampling.
Category: Requirements
What are common risks of Generative AI?
Hallucinations, bias, IP concerns, data leakage, and unsafe outputs. Mitigate with guardrails, evaluation suites, and governance.
Category: Risk
How does RAG improve Generative AI?
RAG retrieves relevant documents to ground outputs, reducing hallucinations and enabling up-to-date or domain-specific responses.
Category: Best practices
What’s the role of fine-tuning in Generative AI?
It adapts base models to domain styles or tasks, improving accuracy and alignment beyond prompt engineering alone.
Category: Example
How do you evaluate Generative AI systems?
Use automated metrics, human review, red-teaming, and task benchmarks; monitor toxicity, factuality, and bias.
Category: Testing
Why is prompt design important in Generative AI?
Prompts condition model behavior; structured instructions, context, and constraints guide quality, safety, and consistency.
Category: Best practices
What governance frameworks apply to Generative AI?
Model cards, risk assessments, EU AI Act categories, and internal policies ensure transparency, compliance, and responsible use.
Category: Requirements
Can Generative AI create synthetic training data?
Yes, to augment datasets or preserve privacy, but quality and bias controls are vital.
Category: Example