AI Visibility Guide
Learn AI Visibility
A structured guide to how AI systems discover, retrieve, select, recommend, and cite information — and what that means for AI Visibility.
01 Part I — Foundations of Large Language Models
01.01 Chapter 1 - From Text to Model Input
- 01.01.01 Chapter Goal
- 01.01.02 Chapter Contents
- 01.01.03 1. Text
- 01.01.04 2. Character
- 01.01.05 3. Word
- 01.01.06 4. Vocabulary
- 01.01.07 5. Token
- 01.01.08 6. Why Not Just Use Words?
- 01.01.09 7. Subword Tokenization
- 01.01.10 8. Common Words vs. Rare Words
- 01.01.11 9. Tokenization
- 01.01.12 10. Tokenizer
- 01.01.13 11. Token ID
- 01.01.14 12. An Important Distinction: ID Is Not Meaning
- 01.01.15 13. Vocabulary Size
- 01.01.16 14. Rare Tokens and Rare Text
- 01.01.17 15. Token Count
- 01.01.18 16. Context
- 01.01.19 17. Context Is Not the Same as Permanent Knowledge
- 01.01.20 18. Context Window, Context Limit, and Context Length
- 01.01.21 19. A Simple Analogy
- 01.01.22 20. Input Context
- 01.01.23 21. Output Tokens
- 01.01.24 22. Input vs. Output Tokens
- 01.01.25 23. Token Budget
- 01.01.26 24. Context Allocation
- 01.01.27 25. Information Competition Inside the Context Window
- 01.01.28 26. Long Context
- 01.01.29 27. Why Tokenization Matters for Long Context
- 01.01.30 28. Why This Matters for Different Languages
- 01.01.31 29. Why This Matters for Names and Brands
- 01.01.32 30. The Complete Pipeline So Far
- 01.01.33 31. What We Have Not Explained Yet
- 01.01.34 32. Mental Model
- 01.01.35 Chapter 1 Recap
- 01.01.36 The Single Most Important Idea from Chapter 1
01.02 Chapter 2- Numerical Representation of Meaning
- 01.02.01 Chapter Goal
- 01.02.02 Chapter Contents
- 01.02.03 1. The Problem: Computers Need Numbers
- 01.02.04 2. Vector Fundamentals
- 01.02.05 3. Embeddings
- 01.02.06 4. Semantic Relationships
- 01.02.07 5. Measuring Similarity
- 01.02.08 6. Important Distinctions
- 01.02.09 7. A Complete Simplified Example
- 01.02.10 8. Connecting the Concepts
- 01.02.11 9. Why These Concepts Matter
- 01.02.12 10. Common Misunderstandings
- 01.02.13 11. How the Concepts Connect
- 01.02.14 Mental Model
- 01.02.15 Chapter 2 Recap
- 01.02.16 The Single Most Important Idea from Chapter 2
01.03 Chapter 3- Transformer Architecture
- 01.03.01 Chapter Goal
- 01.03.02 Chapter Contents
- 01.03.03 1. The Problem: Meaning Depends on Context
- 01.03.04 2. From Static Representation to Contextual Representation
- 01.03.05 3. Language Models and Large Language Models
- 01.03.06 4. Parameters and Model Size
- 01.03.07 5. Neural Networks and Architecture
- 01.03.08 6. The Transformer
- 01.03.09 7. Transformer Layers
- 01.03.10 8. Attention and Self-Attention
- 01.03.11 9. Attention Scores
- 01.03.12 10. Query, Key, and Value
- 01.03.13 11. The Attention Pipeline
- 01.03.14 12. Contextual Representations and Embeddings
- 01.03.15 13. Token Relationships and Context
- 01.03.16 14. Long-Range Dependencies
- 01.03.17 15. Multi-Head Attention
- 01.03.18 16. Repeated Contextualization Across Layers
- 01.03.19 17. A Complete Simplified Transformer Example
- 01.03.20 18. Why These Concepts Matter
- 01.03.21 19. Common Misunderstandings
- 01.03.22 20. How the Concepts Connect
- 01.03.23 Mental Model
- 01.03.24 Chapter 3 Recap
- 01.03.25 The Single Most Important Idea from Chapter 3
01.04 Chapter 4- How LLMs Learn
- 01.04.01 Chapter Goal
- 01.04.02 Chapter Contents
- 01.04.03 1. The Missing Piece: Where Do the Parameters Come From?
- 01.04.04 2. What Is Training?
- 01.04.05 3. Training Data
- 01.04.06 4. The Model Needs a Learning Objective
- 01.04.07 5. Next-Token Prediction
- 01.04.08 6. Prediction Error and Parameter Updates
- 01.04.09 7. The Training Loop
- 01.04.10 8. Why Next-Token Prediction Is More Powerful Than It Sounds
- 01.04.11 9. What Does an LLM Learn?
- 01.04.12 10. Learned Representations
- 01.04.13 11. Pre-Training and Base Models
- 01.04.14 12. Training vs. Inference
- 01.04.15 13. Generalization and Memorization
- 01.04.16 14. What Shapes Learned Associations?
- 01.04.17 15. Training Cutoff
- 01.04.18 16. A Complete Simplified Training Example
- 01.04.19 17. From Training to Learned Behavior
- 01.04.20 18. Training Does Not Guarantee Truth
- 01.04.21 19. Common Misunderstandings
- 01.04.22 20. How the Concepts Connect
- 01.04.23 Mental Model
- 01.04.24 Chapter 4 Recap
- 01.04.25 The Single Most Important Idea from Chapter 4
01.05 Chapter 5- How LLMs Generate Answers
- 01.05.01 Chapter Goal
- 01.05.02 Chapter Contents
- 01.05.03 1. Training Is Finished — Now What?
- 01.05.04 2. Autoregressive Generation
- 01.05.05 3. The Model Does Not Choose from Complete Answers
- 01.05.06 4. Probability Distribution
- 01.05.07 5. Candidate Tokens
- 01.05.08 6. Token Probability
- 01.05.09 7. Token Probabilities Change After Every Generated Token
- 01.05.10 8. Generation as a Path Through Possibilities
- 01.05.11 9. How Is the Next Token Selected?
- 01.05.12 10. What Is Sampling?
- 01.05.13 11. Deterministic vs. Probabilistic Generation
- 01.05.14 12. Why Randomness Can Be Useful
- 01.05.15 13. Temperature
- 01.05.16 14. A Complete Simplified Generation Example
- 01.05.17 15. Generation Depends on the Entire Available Context
- 01.05.18 16. Why the Same Prompt Can Produce Different Answers
- 01.05.19 17. Model Confidence and Uncertainty
- 01.05.20 18. Calibration
- 01.05.21 19. Hallucination, Unsupported Generation, and Fabrication
- 01.05.22 20. Hallucination Cascades
- 01.05.23 21. Generated Answers Can Combine Learned Knowledge and Current Context
- 01.05.24 22. Generation vs. Retrieval and Verification
- 01.05.25 23. The Complete Generation Pipeline
- 01.05.26 24. Common Misunderstandings
- 01.05.27 25. How the Concepts Connect
- 01.05.28 Mental Model
- 01.05.29 Chapter 5 Recap
- 01.05.30 The Single Most Important Idea from Chapter 5
01.06 Chapter 6- Turning a Base Model into an Assistant
- 01.06.01 Chapter Goal
- 01.06.02 Chapter Contents
- 01.06.03 1. The Problem with a Base Model
- 01.06.04 2. Text Completion vs. Instruction Following
- 01.06.05 3. What Is Post-Training?
- 01.06.06 4. Pre-Training vs. Post-Training
- 01.06.07 5. Fine-Tuning
- 01.06.08 6. Domain Fine-Tuning
- 01.06.09 7. Behavioral Fine-Tuning
- 01.06.10 8. Instruction Tuning
- 01.06.11 9. What Is Instruction Following?
- 01.06.12 10. From Instruction Following to Assistant Behavior
- 01.06.13 11. Human Preferences and Preference Data
- 01.06.14 12. RLHF
- 01.06.15 13. Preference Optimization
- 01.06.16 14. How Post-Training Changes Behavior
- 01.06.17 15. What Is a Prompt?
- 01.06.18 16. Prompt vs. Training
- 01.06.19 17. System, Developer, and User Instructions
- 01.06.20 18. Prompt Hierarchy and Instruction Priority
- 01.06.21 19. Prompts and Generation Constraints
- 01.06.22 20. Structured Output
- 01.06.23 21. What Is JSON?
- 01.06.24 22. JSON Schema
- 01.06.25 23. Structured Output vs. Factual Truth
- 01.06.26 24. Post-Training and Prompting Work Together
- 01.06.27 25. Assistant Behavior Is a Layered System
- 01.06.28 26. A Complete Simplified Assistant Example
- 01.06.29 27. Common Misunderstandings
- 01.06.30 28. How the Concepts Connect
- 01.06.31 Mental Model
- 01.06.32 Chapter 6 Recap
- 01.06.33 The Single Most Important Idea from Chapter 6
02 Part II — Information Retrieval
02.01 Chapter 7- Information Retrieval Fundamentals
- 02.01.01 Chapter Goal
- 02.01.02 Chapter Contents
- 02.01.03 1. Why Information Retrieval Exists
- 02.01.04 2. What Is Information Retrieval?
- 02.01.05 3. Retrieval
- 02.01.06 4. Retrieval vs. Generation
- 02.01.07 5. The Retriever
- 02.01.08 6. The Corpus
- 02.01.09 7. Documents and Passages
- 02.01.10 8. What Is a Query?
- 02.01.11 9. Search Query vs. Attention Query
- 02.01.12 10. The Retrieval Problem
- 02.01.13 11. Crawling
- 02.01.14 12. Indexing and the Search Index
- 02.01.15 13. Query-Time Retrieval
- 02.01.16 14. Candidate Generation
- 02.01.17 15. What Is Relevance?
- 02.01.18 16. Relevance Score
- 02.01.19 17. Query–Document and Query–Passage Relevance
- 02.01.20 18. Ranking
- 02.01.21 19. Retrieval vs. Ranking
- 02.01.22 20. Reranking
- 02.01.23 21. Retrieval Can Operate at Multiple Granularities
- 02.01.24 22. Retrieval Is a Selection Problem
- 02.01.25 23. A Complete Simplified Retrieval Example
- 02.01.26 24. Retrieval in a Traditional Search Engine
- 02.01.27 25. Retrieval in an AI System
- 02.01.28 26. Retrieval and Generation Failures
- 02.01.29 27. Candidate Generation Dependency
- 02.01.30 28. Query-Specific Ranking and Intent
- 02.01.31 29. Basic Retrieval Pipeline
- 02.01.32 30. Common Misunderstandings
- 02.01.33 31. How the Concepts Connect
- 02.01.34 Mental Model
- 02.01.35 Chapter 7 Recap
- 02.01.36 The Single Most Important Idea from Chapter 7
02.02 Chapter 8- Lexical and Traditional Search
- 02.02.01 Chapter Goal
- 02.02.02 Chapter Contents
- 02.02.03 1. From Retrieval Architecture to Retrieval Method
- 02.02.04 2. What Is Keyword Search?
- 02.02.05 3. What Is Lexical Search?
- 02.02.06 4. Keyword Search vs. Lexical Search
- 02.02.07 5. Term Matching
- 02.02.08 6. Exact Match
- 02.02.09 7. Exact Phrase Matching
- 02.02.10 8. Why Simple Match Counting Is Not Enough
- 02.02.11 9. Keyword Relevance
- 02.02.12 10. Term Frequency
- 02.02.13 11. Document Frequency
- 02.02.14 12. Term Frequency vs. Document Frequency
- 02.02.15 13. Rare Terms Can Carry More Information
- 02.02.16 14. TF-IDF
- 02.02.17 15. Sparse Retrieval
- 02.02.18 16. What Is a Sparse Vector?
- 02.02.19 17. What Is an Inverted Index?
- 02.02.20 18. Posting Lists
- 02.02.21 19. Inverted Index and Candidate Generation
- 02.02.22 20. Boolean Retrieval
- 02.02.23 21. The Problem of Document Length
- 02.02.24 22. What Is BM25?
- 02.02.25 23. How BM25 Behaves
- 02.02.26 24. Why Lexical Search Is Powerful
- 02.02.27 25. The Central Weakness: Vocabulary Mismatch
- 02.02.28 26. Synonyms
- 02.02.29 27. Exact Words vs. Meaning
- 02.02.30 28. Search Intent Mismatch
- 02.02.31 29. Lexical vs. Semantic Strengths
- 02.02.32 30. Lexical Search and Candidate Generation
- 02.02.33 31. Lexical Search in RAG
- 02.02.34 32. Lexical Search in Web Search
- 02.02.35 33. A Complete Simplified Lexical Retrieval Example
- 02.02.36 34. Traditional Search Is Statistical, Not Truly “Understanding” the Query
- 02.02.37 35. Common Misunderstandings
- 02.02.38 36. How the Concepts Connect
- 02.02.39 Mental Model
- 02.02.40 Chapter 8 Recap
- 02.02.41 The Single Most Important Idea from Chapter 8
02.03 Chapter 9- Dense and Semantic Retrieval
- 02.03.01 Chapter Goal
- 02.03.02 Chapter Contents
- 02.03.03 1. From Words to Meaning
- 02.03.04 2. What Is Dense Retrieval?
- 02.03.05 3. What Is a Dense Vector?
- 02.03.06 4. Sparse Vector vs. Dense Vector
- 02.03.07 5. Dense Does Not Mean “More Accurate”
- 02.03.08 6. What Is Semantic Retrieval?
- 02.03.09 7. Dense Retrieval and Semantic Retrieval Are Related but Not Identical
- 02.03.10 8. Returning to Embeddings
- 02.03.11 9. Token Embeddings vs. Text Embeddings
- 02.03.12 10. Embedding Is a Representation, Not the Retrieval Process
- 02.03.13 11. What Is an Embedding Model?
- 02.03.14 12. Embedding Space
- 02.03.15 13. Document Embeddings
- 02.03.16 14. Query Embeddings
- 02.03.17 15. Query and Documents Need Comparable Representations
- 02.03.18 16. A Simplified Dense Retrieval Example
- 02.03.19 17. What Is Vector Similarity?
- 02.03.20 18. Cosine Similarity
- 02.03.21 19. Dot Product
- 02.03.22 20. Euclidean Distance
- 02.03.23 21. What Is Nearest-Neighbor Search?
- 02.03.24 22. Top-K Retrieval
- 02.03.25 23. Vector Index
- 02.03.26 24. Approximate Nearest-Neighbor Search
- 02.03.27 25. Dense Retrieval as Candidate Generation
- 02.03.28 26. A Typical Dense Retrieval Preparation Stage
- 02.03.29 27. A Typical Dense Retrieval Query-Time Stage
- 02.03.30 28. Bi-Encoder Retrieval
- 02.03.31 29. Query–Document Similarity
- 02.03.32 30. Vocabulary Mismatch and Paraphrases
- 02.03.33 31. Limits of Semantic Similarity
- 02.03.34 32. Entity and Identifier Problems
- 02.03.35 33. Embedding and Retrieval Quality
- 02.03.36 34. Passage-Level Dense Retrieval
- 02.03.37 35. A Vector Database
- 02.03.38 36. Metadata Filtering
- 02.03.39 37. Dense Retrieval and Ranking
- 02.03.40 38. Dense Retrieval vs. Reranking
- 02.03.41 39. Dense Retrieval vs. LLM Generation
- 02.03.42 40. Semantic Search Is Not “Search the LLM's Memory”
- 02.03.43 41. Lexical Search and Dense Search Use Different Evidence
- 02.03.44 42. Dense Retrieval Strengths and Limitations
- 02.03.45 43. Dense Retrieval and Information Competition
- 02.03.46 44. A Complete Simplified Dense Retrieval Example
- 02.03.47 45. The Same Example with Lexical Retrieval
- 02.03.48 46. Why Hybrid Retrieval Becomes Natural
- 02.03.49 47. Common Misunderstandings
- 02.03.50 48. How the Concepts Connect
- 02.03.51 Mental Model
- 02.03.52 Chapter 9 Recap
- 02.03.53 The Single Most Important Idea from Chapter 9
02.04 Chapter 10- Hybrid Retrieval
- 02.04.01 Chapter Goal
- 02.04.02 Chapter Contents
- 02.04.03 1. Why Hybrid Retrieval Exists
- 02.04.04 2. What Is Hybrid Retrieval?
- 02.04.05 3. Hybrid Retrieval Is Not a New Representation Type
- 02.04.06 4. The Two Signals
- 02.04.07 5. Complementary Failure Modes
- 02.04.08 6. Parallel Hybrid Retrieval
- 02.04.09 7. What Is Fusion?
- 02.04.10 8. Candidate Union
- 02.04.11 9. Why Candidate Union Can Improve Recall
- 02.04.12 10. Candidate Intersection
- 02.04.13 11. Why Intersection Can Be Too Strict
- 02.04.14 12. Candidate Generation vs. Fusion Ranking
- 02.04.15 13. The Score Problem
- 02.04.16 14. Retrieval Scores Are Not Directly Comparable
- 02.04.17 15. Score Normalization
- 02.04.18 16. Weighted Score Fusion
- 02.04.19 17. Why Use Different Weights?
- 02.04.20 18. Query-Dependent Weighting
- 02.04.21 19. Score Fusion vs. Rank Fusion
- 02.04.22 20. Why Rank Fusion Can Be Useful
- 02.04.23 21. Reciprocal Rank Fusion
- 02.04.24 22. RRF Intuition
- 02.04.25 23. RRF Does Not Need Raw Score Normalization
- 02.04.26 24. RRF Does Not Mean Every Retriever Has Equal Quality
- 02.04.27 25. Hybrid Retrieval and Recall
- 02.04.28 26. Higher Recall Can Increase Noise
- 02.04.29 27. Hybrid Retrieval Does Not Replace Ranking
- 02.04.30 28. Hybrid Retrieval and Reranking
- 02.04.31 29. Hybrid Retrieval Can Work at Passage Level
- 02.04.32 30. Example: Hybrid Retrieval in a Support Knowledge Base
- 02.04.33 31. Example: Hybrid Retrieval with an Identifier
- 02.04.34 32. Metadata Still Matters
- 02.04.35 33. Filtering Before Retrieval
- 02.04.36 34. Filtering After Retrieval
- 02.04.37 35. Hybrid Retrieval Is Not Just “BM25 + Embeddings”
- 02.04.38 36. Query Routing vs. Hybrid Retrieval
- 02.04.39 37. Hybrid Retrieval, Generation, and Truth
- 02.04.40 38. A Complete Simplified Hybrid Retrieval Example
- 02.04.41 39. When Hybrid Retrieval Helps Most
- 02.04.42 40. When Hybrid Retrieval May Be Unnecessary
- 02.04.43 41. Cost and Complexity
- 02.04.44 42. Hybrid Retrieval Must Be Evaluated
- 02.04.45 43. Common Misunderstandings
- 02.04.46 44. How the Concepts Connect
- 02.04.47 Mental Model
- 02.04.48 Chapter 10 Recap
- 02.04.49 The Single Most Important Idea from Chapter 10
02.05 Chapter 11- Ranking and Reranking
- 02.05.01 Chapter Goal
- 02.05.02 Chapter Contents
- 02.05.03 1. Candidate Generation Is Not the Final Decision
- 02.05.04 2. What Is Ranking?
- 02.05.05 3. What Is a Ranker?
- 02.05.06 4. Ranking Score
- 02.05.07 5. Retrieval vs. Ranking
- 02.05.08 6. Retrieval Often Already Produces an Initial Ranking
- 02.05.09 7. First-Stage Ranking
- 02.05.10 8. Why Multi-Stage Ranking Exists
- 02.05.11 9. What Is Reranking?
- 02.05.12 10. Why Reranking Can Be Better
- 02.05.13 11. Bi-Encoder Review
- 02.05.14 12. The Limitation of Separate Encoding
- 02.05.15 13. What Is a Cross-Encoder?
- 02.05.16 14. Bi-Encoder vs. Cross-Encoder
- 02.05.17 15. Why Cross-Encoders Are More Expensive
- 02.05.18 16. Joint Query–Document Interaction
- 02.05.19 17. Reranking Example
- 02.05.20 18. Reranking Does Not Usually Search the Whole Corpus
- 02.05.21 19. Retrieval Recall and Ranking Precision
- 02.05.22 20. Ranking Can Use More Than Text Similarity
- 02.05.23 21. Ranking Features
- 02.05.24 22. Retrieval Score vs. Ranking Features
- 02.05.25 23. What Is Learning to Rank?
- 02.05.26 24. Where Do Ranking Labels Come From?
- 02.05.27 25. Pointwise Ranking
- 02.05.28 26. Pairwise Ranking
- 02.05.29 27. Listwise Ranking
- 02.05.30 28. Cross-Encoder Reranking vs. Learning to Rank
- 02.05.31 29. Reranking After Hybrid Retrieval
- 02.05.32 30. Why Top Results Matter So Much
- 02.05.33 31. Ranking and the Context Window
- 02.05.34 32. Ranking Errors Can Become Generation Errors
- 02.05.35 33. Relevance Is Not the Only Ranking Objective
- 02.05.36 34. Relevance vs. Authority
- 02.05.37 35. Relevance vs. Freshness
- 02.05.38 36. Ranking vs. Factual Verification
- 02.05.39 37. Diversity in Ranking
- 02.05.40 38. Reranking Can Be Query-Aware in Greater Detail
- 02.05.41 39. Reranking Is Still Not Reasoning Over Everything
- 02.05.42 40. LLM-Based Reranking
- 02.05.43 41. A Complete Simplified Ranking Pipeline
- 02.05.44 42. Ranking in Traditional Search vs. AI Search
- 02.05.45 43. Ranking and AI Visibility
- 02.05.46 44. Common Misunderstandings
- 02.05.47 45. How the Concepts Connect
- 02.05.48 Mental Model
- 02.05.49 Chapter 11 Recap
- 02.05.50 The Single Most Important Idea from Chapter 11
02.06 Chapter 12- Measuring Retrieval Quality
- 02.06.01 Chapter Goal
- 02.06.02 Chapter Contents
- 02.06.03 1. What Is Retrieval Evaluation?
- 02.06.04 2. Evaluation Requires a Reference
- 02.06.05 3. What Is Ground Truth?
- 02.06.06 4. Relevance Judgments
- 02.06.07 5. Binary vs. Graded Relevance
- 02.06.08 6. Precision
- 02.06.09 7. The Intuition Behind Precision
- 02.06.10 8. Recall
- 02.06.11 9. Precision vs. Recall
- 02.06.12 10. The Precision–Recall Trade-Off
- 02.06.13 11. Why We Need @K Metrics
- 02.06.14 12. What Does K Mean?
- 02.06.15 13. Precision@K
- 02.06.16 14. Recall@K
- 02.06.17 15. Candidate Recall
- 02.06.18 16. Precision Matters More Near the End
- 02.06.19 17. Hit Rate@K
- 02.06.20 18. When Hit Rate Is Useful
- 02.06.21 19. Reciprocal Rank
- 02.06.22 20. Mean Reciprocal Rank — MRR
- 02.06.23 21. What MRR Measures Well
- 02.06.24 22. Why Position Matters in Ranking Evaluation
- 02.06.25 23. Discounted Cumulative Gain — DCG
- 02.06.26 24. Why DCG Uses Graded Relevance
- 02.06.27 25. NDCG — Normalized Discounted Cumulative Gain
- 02.06.28 26. NDCG Example
- 02.06.29 27. NDCG@K
- 02.06.30 28. Average Precision
- 02.06.31 29. Mean Average Precision — MAP
- 02.06.32 30. MRR vs. MAP vs. NDCG
- 02.06.33 31. Precision@K vs. NDCG@K
- 02.06.34 32. Metrics Depend on the Task
- 02.06.35 33. Retrieval Evaluation for Multi-Stage Systems
- 02.06.36 34. Evaluation of a Reranker
- 02.06.37 35. Evaluation of Candidate Generation
- 02.06.38 36. What Is a Test Query Set?
- 02.06.39 37. Query Diversity Matters
- 02.06.40 38. Offline Evaluation
- 02.06.41 39. Online Evaluation
- 02.06.42 40. Clicks Are Not Pure Relevance Labels
- 02.06.43 41. Offline Metrics vs. Online Success
- 02.06.44 42. Evaluation Metrics Are Not Product Goals
- 02.06.45 43. Retrieval Evaluation in RAG
- 02.06.46 44. Context Recall
- 02.06.47 45. Context Precision
- 02.06.48 46. Retrieval Metrics and AI Visibility
- 02.06.49 47. Retrieval Presence vs. Final Answer Presence
- 02.06.50 48. No Single Metric Tells the Whole Story
- 02.06.51 49. A Practical Metric Stack
- 02.06.52 50. A Complete Simplified Evaluation Example
- 02.06.53 51. Common Misunderstandings
- 02.06.54 52. How the Concepts Connect
- 02.06.55 Mental Model
- 02.06.56 Chapter 12 Recap
- 02.06.57 The Single Most Important Idea from Chapter 12