AI Foundations for Everyone
Everything you need to understand modern AI in one course: how models are built and trained, the transformer machinery inside them, where they fail, and how to prompt, verify, delegate, and stay current, with the landmark research taught along the way.
20 sections ยท 85 lessons
Course outline
Language Models Basics
- Inside ChatGPT
- How Prompts Work
- Probability & Randomness
- Temperature
How Models Learn
- Training Data Scale
- Generalization
- Parameters
- Tuning the Model
Training in Practice
- The Training Process
- Developing Understanding
- Scale of Computation
- Hardware Reality
Making AI Helpful
- Pre-training Basics
- Supervised Fine-Tuning
- Reinforcement Learning
- Complete Training Pipeline
- GPU Technology
- The Parallelization Problem
- Why It Stops at a Date
Transformer Innovation
- Introduction to Transformers
- Words to Numbers
- Understanding Embeddings
- Introduction to Attention
- How Attention Works
Connecting the Pieces
- Feed-Forward Networks
- Multi-Layer Basics
- From Vectors to Predictions
- Attention Is All You Need
- Complete Pipeline
AI Strengths & Limitations
- What LLMs Excel At
- Next to Master
- Core LLM Limitations
- Structural Constraints
Practical AI Use
- Prompt Engineering Fundamentals
- Advanced Prompting Techniques
- AI Literacy: Engaging & Creating
- AI Literacy: Managing & Designing
- Your AI-Augmented Future
Memory & Context
- The Context Window
- Lost in the Middle
- Context Engineering
Reasoning Models
- Test-Time Compute
- When Thinking Pays Off
- The Reasoning Race
The Jagged Frontier
- The Boundary Is Not Where You Think
- Sort Your Own Tasks
- Centaurs and Cyborgs
Hallucination & Fabrication
- Why Models Hallucinate
- Where To Expect It
- The Arithmetic of Bluffing
- It Was Trained To Please You
Checking the Work
- Numbers Are Just Text To It
- Make It Show Its Working
- Reading Well Is Not Being Right
- Match The Checking To The Stakes
- How To Get A Real Critique
Documents & Retrieval
- Retrieval-Augmented Generation
- It Only Read Some Of It
- Grounded Is Not Guaranteed
- The Shape Is Part Of The Request
- Structure Makes Checking Cheap
Choosing Your Model
- Not A Ladder
- The Surface Matters More
- Benchmarks Are Not Your Task
- The Open-Weights Movement
Beyond Text
- Reading And Hearing Are Predictions Too
- When The Media Itself Is Made Up
- Generating Images and Video
AI Agents
- The Loop
- Tools and MCP
- Errors Compound
- Where The Brakes Go
Delegation & Collaboration
- Parts, Not Whole Tasks
- What Never Gets Handed Over
- The Workslop Study
- Feeling Faster, Measured Slower
Safety, Privacy & Rules
- Prompt Injection
- Risk Follows Reach
- Where Your Data Goes
- Retention Is A Separate Question
- The Defaults You Did Not Choose
- Disclosure And The Rules
Staying Current
- What Churns And What Holds
- A Routine You Will Actually Run
- The Road From Here