AI vs ML vs LLM vs GenAI: the terms, in plain English.

Four words get used in the same sentence as if they meant the same thing. They do not. The distinctions matter when you are buying a tool, evaluating a vendor, or trying to scope a pilot.

Artificial intelligence (AI)

The broadest term. Any system that does something that, twenty years ago, would have looked like it required human intelligence. Includes everything below. Includes things that are not below: rule-based systems, search engines that rank by relevance, recommendation algorithms, computer vision systems, autonomous-driving stacks. The word "AI" by itself tells you almost nothing about what is inside.

Machine learning (ML)

A subset of AI. A system that learns patterns from data, instead of being programmed with explicit rules. Includes everything from a credit-scoring model that predicts default risk to a fraud-detection system to an image classifier. ML predates the recent LLM boom by decades. When someone says "we have AI in our product," ninety percent of the time they mean ML.

Large language model (LLM)

A specific architecture inside ML, built on transformer networks trained on enormous amounts of text. Examples: Claude, GPT, Gemini, Llama. An LLM is a thing. It is the underlying engine. You access it through an API or through a product wrapped around it.

Generative AI (GenAI)

A category of capability: AI systems that produce new content rather than classifying or scoring existing content. Generative AI for text typically uses LLMs. Generative AI for images uses different model families (diffusion models). Generative AI for code is typically an LLM trained heavily on code. GenAI is the capability. The model is the engine.

A practical example

When you talk to ChatGPT, you are using:

- A product (ChatGPT) - Built on a large language model (GPT) - Which is a type of machine learning system - Which is a kind of artificial intelligence - Used in a generative AI mode

All four words apply at the same time. The distinctions matter when a vendor says "we have AI" and you are trying to figure out what they actually built.

What to ask vendors

When evaluating an AI product, the questions are:

1. What underlying model? (Specific name and version.) 2. Hosted where? (Vendor cloud, your cloud, on-prem.) 3. Trained on what? (Public data, your data, customer data they sell to others.) 4. What is the actual task it does? (Classification, generation, retrieval, scoring.)

If a vendor cannot answer those four questions, they probably do not have a coherent product yet.

The LearnTrainAI workshop covers this evaluation framework explicitly in the procurement-readiness module.