Understanding AI & Generative AI
Give executives a technically accurate mental model of modern AI without turning the course into an engineering program.
Executives do not need to build AI, but they need a model of it accurate enough to judge claims, set expectations and see risk. AI is a label for a family of capabilities, and the label says little on its own. The useful questions are what a system does, how it was built, how reliably it performs and what happens when it is wrong. Automation follows rules people wrote; machine learning follows patterns learned from data, and those patterns hold only while new cases resemble the ones it learned from.
Modern generative AI rests on foundation models, trained broadly once and adapted to many tasks. A large language model predicts the next token from its context, which makes it fluent across a wide range of work but leaves it with patterns rather than records. Fluent output is likely, not verified. Two consequences follow for leaders. What the model can see at the moment of the question decides much of the quality, so context and retrieval from the organization's own systems usually matter more than retraining. And the cost and limits of running a model, not only of training it, shape what is affordable.
The module ends where AI moves from answering to acting: an agent pursues a goal through steps, tools and checks, and small misses compound across the whole job. The AI Revolution explained why this technology matters; Business Value of AI, which follows, asks what it is worth and how to prove it.
Questions this module answers
- What does a system labeled AI actually do, and how should its claims be tested?
- How do machine learning and deep learning learn, and why do their patterns stop holding?
- What are foundation models and large language models, and what do they not know?
- Why does what the model sees at question time matter as much as the model itself?
- What makes an agent different from an assistant, and how should it be judged?
The chapters
AI foundations
Defines AI by capability, separates it from automation and machine learning and sorts the labels used to describe it.
- 001 What Exactly Is Artificial Intelligence? AI is a label for a family of capabilities; translate it into what a system does, how reliably, and what happens when it is wrong. 13 min read
- 002 AI vs Automation Automation follows rules people wrote; AI follows patterns learned from data. They fail differently, so combine them deliberately, starting from the simplest design. 13 min read
- 003 AI vs Machine Learning AI is the field; machine learning is one way to build it. Ask which parts learn, how, and where their answer key comes from. 11 min read
- 004 The Major Types of AI AI labels answer four different questions: what a system does, how it is built, how general it is and how much it does alone. 12 min read
How modern AI works
Explains how models learn, what deep learning adds, how training differs from inference and how data, models and compute set the ceiling.
- 001 How Machine Learning Learns A model learns patterns by reducing its error on examples with known answers; those patterns pay only if they hold on unseen cases and keep holding as the world moves. 13 min read
- 002 Deep Learning — The Executive Mental Model Deep learning is machine learning with many-layered neural networks that learn their own features from raw data. Depth is capacity, not a guarantee. 11 min read
- 003 Training vs Inference Training creates a model's capability; inference delivers it on every request, leaves the model unchanged and carries much of the running cost. 11 min read
- 004 Data, Models and Compute AI capability comes from data, models and compute working like factors in a product, so the weakest of the three sets the ceiling. 11 min read
Foundation models
Shows how foundation and language models are built and adapted, how tokens, context and embeddings work and when other modalities add information.
- 001 Foundation Models A foundation model is trained broadly once, then adapted to many tasks. It is only the bottom layer, and everything built on it inherits its flaws. 11 min read
- 002 Large Language Models An LLM predicts the next token from context. That yields a broad language capability, but what it learned is patterns, not records, so facts must come from systems of record. 12 min read
- 003 Tokens, Context and Embeddings Tokens are what a model reads and bills, context is what it sees, embeddings find content by meaning: send the right information, not all of it. 12 min read
- 004 Multimodal AI Multimodal AI reads and produces images, audio and video as well as text; add a modality only when it holds information that changes the decision. 12 min read
Generative AI and agents
Explains how generation works, how prompting, context and retrieval shape answers, and how agents pursue goals through steps and tools.
- 001 How Generative AI Works Generative AI composes new output, token by token or by refining noise, drawing on chance at every step; fluent output is likely, not verified. 12 min read
- 002 Prompting and Context Engineering — Executive Mental Model Prompting tells the model what to do; context engineering decides what it has to work with. Many wrong answers come from what the model could not see. 11 min read
- 003 RAG and Enterprise Knowledge — Executive Mental Model Retrieve, don't retrain: keep enterprise knowledge in its home systems and hand the model only what is relevant, current and permitted, at question time. 11 min read
- 004 From AI Assistants to AI Agents An assistant answers a request; an agent pursues a goal through a loop of steps, tools and checks. Judge it on the whole job. 11 min read
After this module you can
- Translate an AI label into a capability profile: what the system does, how reliably and what happens when it is wrong.
- Explain why learned patterns can fail on new cases and drift as the world changes.
- Distinguish what a language model learned in training from the information it is given at question time.
- Choose between prompting, retrieval and retraining for a given business need.
- Judge an agent on the reliability of the whole job rather than on its best answer.