Module 3.1 · Topic 1

What Is Generative AI?

Bottom Line Up Front: Generative AI is a subset of machine learning that creates new content by predicting what comes next based on training data patterns. Unlike traditional software that executes predetermined rules,…

1.1 The AI Hierarchy: Artificial Intelligence, Machine Learning, and Generative AI

These three categories form a spectrum of increasing specificity. Knowing where a tool sits tells you immediately whether outputs are variable (probabilistic) or deterministic.

Artificial Intelligence (AI): Any system that mimics intelligence by perceiving, reasoning, or learning from data.
Machine Learning (ML): A subset of AI where systems improve by learning patterns from data.
Generative AI (GenAI): A subset of ML that generates content by predicting what comes next. Trained to reproduce patterns.

1.2 Essential Terminology for the AI Era

Large Language Model (LLM): A generative AI trained on vast text data to predict the next word. The "large" refers to scale, not capability.
Token: A unit of text (roughly a word) that AI processes. Longer inputs consume more tokens.
Hallucination: When AI generates plausible but false information with confidence. It's pattern-matching that produces invented facts.
Fine-tuning: Adapting a model to a specific task by training on specialized data. Better on narrow tasks, may lose general knowledge.
Knowledge Cutoff: The training date boundary. An AI trained through February 2026 cannot (without incorporate web or other resources) answer questions about events after.
Temperature: A setting controlling variation. Higher = more creative. Lower = more consistent. In most end-user applications, temperature settings are decided by the developer and are not controllable at the user level.

1.3 What Makes Generative AI Different From Traditional Software

Dimension	Generative AI	Traditional Software
How it works	Predicts next output based on training patterns. Outputs probabilistic.	Executes predetermined logic. Outputs deterministic.
Consistency	Same input may produce different outputs (inherent).	Same input produces identical output.
Transparency	Difficult to explain output. Black box.	Logic is traceable and explainable.
Factual reliability	Variable. May confidently produce false information.	Reliable for data retrieval.
Failure modes	Confident hallucinations, knowledge gaps, pattern errors.	Logic errors, crashes, edge case misses.

1.4 Core Capabilities and Fundamental Limitations

Capability / Limitation	AI Can	AI Cannot
Text synthesis	Summarize, write, rewrite, outline.	Guarantee factual accuracy. May insert false facts.
Pattern matching	Extract information, identify inconsistencies, categorize.	Understand context as humans do. May miss subtle meanings.
Explanation	Clarify concepts, explain "why," translate jargon.	Know if explanations are current. Based on training data.
Real-time knowledge	Access training data (up to knowledge cutoff).	Answer about events after cutoff or proprietary information.

Topic 1 details

Topic overview

Bottom Line Up Front: Generative AI is a subset of machine learning that creates new content by predicting what comes next based on training data patterns. Unlike traditional software that executes predetermined rules, it produces variable outputs through statistical probability, making it powerful for synthesis tasks but fundamentally unreliable for factual recall.

You've prompted AI effectively. Now learn what generative AI actually is, how it fits within the broader AI landscape, what terminology matters, how it differs from traditional software, and where its real capabilities and hard limits lie.

Topic details

Key takeaway

Knowledge check

What Is Generative AI?: A colleague says generative AI is "just like normal software, only smarter." Using how it produces output, explain one reason the same prompt to a generative model can return a different answer each time, when traditional software cannot.

Show answer

Traditional software executes predetermined logic, so identical input yields identical, deterministic output. Generative AI instead predicts the next token from statistical patterns in its training data and samples from a probability distribution, so the same input can produce different outputs. This variability is inherent to how it works, not a malfunction.