AI FOUNDATIONS
AI Foundations: From Using the Tools to Actually Understanding Them
A 14-chapter path from zero: why the model answers, why it forgets, why it makes things up, and how to pick the right tool
You may already use ChatGPT, Claude, or Gemini, yet still find it hard to judge what tokens, context windows, reasoning models, and embeddings actually mean. This course does not ask you to memorize jargon and assumes no programming background. It uses concrete examples to connect the most important ideas in modern AI into one mental map you can keep using.
Learning path
These 14 chapters move through four steps
The course starts with how a single answer gets made, then closes the three gaps that frustrate people most — memory, freshness, and fabrication — before moving on to how AI represents text and images, and finally back to the decisions you actually make each day: which model to pick, and where to spend your money.
- 01First, how one answer gets made
Understand the next-word prediction at the core of an LLM, how that ability was trained, and how tokens drive both speed and cost.
- 02Then, its three hard limits
The context window sets how much it remembers, the knowledge cutoff sets how much it knows, and hallucination sets where it goes wrong — plus one boundary you have to enforce yourself: privacy.
- 03Next, how it perceives the world
Embeddings and multimodality explain how text becomes vectors and how images and audio enter the model — and therefore why knowledge-base retrieval works at all.
- 04Finally, your own selection criteria
Read the model picker with confidence, know when an agent truly checks a source, interpret parameter counts, and weigh subscriptions against APIs and self-hosted open models.
Learning path
PART 1
Part 1: How a Single Answer Gets Made
Start from the one thing the model actually does — guess the next word — then see how that ability was trained, and why every answer carries a measurable cost.
PART 2
Part 2: What It Remembers, What It Knows, Where It Fails
This part addresses the three most common frustrations in daily use — why it forgets, why it misses recent events, and why it invents things confidently — then closes with the privacy boundary most beginners overlook.
PART 3
Part 3: How It Makes Sense of the World
How text becomes computable coordinates, and how images and audio enter the same model — the shared mechanism behind knowledge-base retrieval and snapping a photo to ask about it.
PART 4
Part 4: Choosing and Controlling Your Tools
Back to the moment you open the tool: what the options in the model picker actually differ on, when the AI really goes and checks instead of guessing, what those parameter counts mean, and whether you should be paying for a subscription or an API.
APPENDIX
Appendix: Advanced Reading and Quick Reference
Appendix A is for readers who tune API parameters and can be skipped by everyone else; Appendix B collects every term in the course onto one page you can come back to.
What you will be able to do
You will not become a machine-learning engineer, but you should be able to explain the core concepts in your own words and apply them to real decisions about tools and problems.
- Explain how a large language model produces an answer, and how that ability comes from pre-training, fine-tuning, and RLHF.
- Connect tokens, context windows, and cost — and know why long conversations drift and how to split long documents.
- Recognize where hallucinations cluster, which answers must be cross-checked, and how to phrase prompts that reduce fabrication.
- Judge what should never be pasted into a chat box, and read privacy settings such as "use my data for training" for what they actually mean.
- Describe what embeddings, multimodality, training data, and knowledge cutoffs each solve.
- Choose between a reasoning model and a fast model for a given task, and know when the AI needs to search or call a tool.
- Interpret parameter counts like 7B / 70B / 671B, and choose among subscription, API, and self-hosted open models on cost, privacy, and control.