Appendix: Common Traps (Master List)

<details> <summary><strong>Chapter 0: What AI Actually Is</strong></summary>

• Treating AI outputs as truth
• Assuming AI understands context
• "It works on my test set, ship it!"
• Anthropomorphizing the model

</details> <details> <summary><strong>Chapter 1: Python & Data</strong></summary>

• Not looking at your data
• Trusting data providers
• Ignoring missing data patterns
• Not versioning data

</details> <details> <summary><strong>Chapter 2: Math You Can't Escape</strong></summary>

• Memorizing formulas without understanding
• Getting stuck in math rabbit holes
• Skipping linear algebra
• Treating probability as just counting

</details> <details> <summary><strong>Chapter 3: Classical ML</strong></summary>

• Not using cross-validation
• Tuning hyperparameters on the test set
• Ignoring class imbalance
• Forgetting about feature scaling

</details> <details> <summary><strong>Chapter 4: Neural Networks</strong></summary>

• Not normalizing inputs
• Using sigmoid for hidden layers
• Not shuffling data
• Forgetting to set model to eval mode
• Not checking for NaNs

</details> <details> <summary><strong>Chapter 5: Transformers & LLMs</strong></summary>

• Trusting LLM outputs without verification
• Using LLMs for tasks requiring reasoning
• Ignoring cost
• Not handling edge cases

</details> <details> <summary><strong>Chapter 6: Modern AI Systems</strong></summary>

• Over-relying on LLMs
• Not versioning prompts
• Ignoring latency
• No fallback logic

</details> <details> <summary><strong>Chapter 7: Production AI</strong></summary>

• Deploying and forgetting
• Optimizing for accuracy alone
• Not planning for retraining
• Adding AI because it's trendy

</details>

---