MODULE 02 / CONCEPTS + EXPERIMENTS
AI Foundations
Learn how intelligent systems work, one concept and one experiment at a time. Start with rules and learning, then build toward models, language, and agents.
Start reading ↗STUDY LIBRARY / NEWEST FIRST
Build your understanding
Browse the latest sections first. Use Start reading and each article's previous and next links to follow the teaching sequence. Each section includes prerequisites, worked examples, and exercises with answer guidance.
Building Good Experimental Habits
From Real-World Problems to AI Tasks
Case Study: From a Vague Request to a Verifiable AI Task
Turn a troubleshooting-assistant proposal into a bounded document recommendation task, with explicit evidence, a baseline, permissions, and testable acceptance criteria.
Business Metrics and Model Metrics
Connect prediction quality to service outcomes, compare errors under explicit costs and capacity limits, and measure whether a model improves the work it supports.
When You Don't Need Machine Learning
Recognize tasks that are better served by rules, queries, search, optimization, or human review, and compare the evidence and operating costs before adding a model.
Detection, Segmentation, Generation, and Decision-Making
Follow a parcel inspection from bounding boxes and pixel masks to a report draft and a permitted action, with worked overlap metrics and expected-loss calculations.
Classification, Regression, Ranking, and Clustering
Choose an AI task by its required output. Work through ticket routing, handling-time estimates, review priorities, and similarity groups with a verified NumPy example.
Defining Goals, Inputs, Outputs, and Constraints
Turn a delivery problem into a testable AI task: define the decision, available inputs, prediction target, operating limits, and evidence needed before release.
Introduction to AI
CPUs, GPUs, GPU Memory, and Compute Costs
Calculate model weights, KV cache, and serving costs, and learn how CPU work, GPU execution, and memory limits shape an AI application.
AI Capabilities and Their Limits
Learn which tasks AI can support, where information and evaluation set limits, and how to measure the cost of mistakes and human review.
Training, Inference, and Deployment
Follow a fitted email classifier from training data to a versioned release, and learn how prediction, evaluation, and deployment differ.
Large Language Models and Agents
Follow an order-support request from token prediction to a bounded tool-calling loop, with a runnable Python example and explicit permission checks.
Discriminative and Generative Models
Learn what each model represents, derive a generative classifier step by step, and verify the probabilities with NumPy.
How AI, Machine Learning, and Deep Learning Fit Together
Understand the relationship between AI, machine learning, and deep learning through one concrete email-filtering system.
From Rule-Based Systems to Machine Learning
Learn what a model learns, how training differs from prediction, and why lower training error needs an independent test.