Unit 1 Review — Preparing for the Quiz
This lesson summarizes everything you've learned in Unit 1 of the AI Foundations course. Review it carefully before taking the quiz.
Summary of the Previous Seven Lessons
Lesson 1: What is Artificial Intelligence?
- AI = simulating human intelligence with machines
- Three types: Narrow AI (exists now), AGI (theoretical), ASI (future concept)
- 3 factors that exploded AI: data + computing + Transformer algorithm
- Core difference: learning from data ≠ manually writing rules
Lesson 2: History of AI
- 1950: Turing asks "Can machines think?" + 1956: Dartmouth Conference (birth of the term)
- Two AI winters: 1974-1980 and 1987-1993 (failed due to resource limitations)
- 2012: AlexNet — the deep learning revolution (ImageNet competition)
- 2017: Transformer paper — architectural parent of every modern LLM
- 2022: ChatGPT — 100 million users in two weeks
Lesson 3: AI vs ML vs Deep Learning
- Nested circles: AI ⊃ ML ⊃ Deep Learning
- Every Deep Learning is ML, every ML is AI — but not the reverse
- ChatGPT and Claude: Deep Learning → ML → AI
Lesson 4: How Does an LLM Work?
- Core idea: predicts the best next word repeatedly
- 3 build stages: Tokenization → Pre-training → Fine-tuning/RLHF
- Transformer + Attention mechanism = connects distant words
- Context Window = short-term memory for each session
Lesson 5: Introduction to LLMs
- Parameters: numerical weights = size metric (but not everything)
- Three big limitations: hallucination + knowledge cutoff + context window
- 2025 models: GPT-4o, Claude 4, Gemini 2.5, Llama 3.1
Lesson 6: AI Tools Tour
- Writing and conversation: ChatGPT / Claude / Gemini / Perplexity
- Images: Midjourney / DALL-E 3 / Adobe Firefly
- Code: GitHub Copilot / Cursor / Claude
- Audio: ElevenLabs / Whisper + Video: RunwayML / Sora
Lesson 7: Chatbot Project
- Building a chatbot = good System Prompt + testing + improvement
- Project phases: define goal → System Prompt → test → refine
Key Terms for the Quiz
| Term | Brief Definition |
|------|-----------------|
| AI | Simulating human intelligence with machines |
| ML (Machine Learning) | AI approach that learns from data |
| Deep Learning | ML using multi-layer neural networks |
| LLM | Large language model trained on trillions of words |
| Token | The basic text unit in LLM models |
| Context Window | Maximum text capacity in a single session |
| Hallucination | Model inventing information that sounds correct but isn't |
| Transformer | Architectural foundation of every modern LLM (2017) |
| Fine-tuning | Adjusting a model after initial training for specific behavior |
| System Prompt | Hidden instructions defining AI identity and behavior |
| Parameters | Numerical weights inside the neural network |
| Knowledge Cutoff | Date after which the model is unaware of events |
Common Mistakes to Avoid
❌ "AI knows everything" — No, it has a knowledge cutoff and sometimes hallucinates
❌ "Every ML is Deep Learning" — Deep Learning is a subset of ML, not the reverse
❌ "More parameters always = better model" — Training efficiency and data quality also matter
❌ "LLM thinks like a human" — It calculates next-word probabilities, it doesn't think
❌ "The first AI winter was in the 2000s" — It was 1974-1980 (before the internet existed)
Pre-Quiz Tip
The quiz tests understanding, not memorization. If you can explain each term in the table to someone who knows nothing about AI, you'll answer every question confidently.
The important sequence to remember:
Transformer (2017) → Pre-training → Fine-tuning + RLHF → LLM ready to use
Good luck! 🎯