Mental Models Cheat Sheet

Information pipeline

Input → Transform → Output → Feedback

Use when debugging any system: Where did bad data enter? Where did the transform fail? Who checks the output?

Compression lens

Complex ideas become usable when you find the smallest accurate picture. If you cannot draw it in 30 seconds, you do not understand it yet.

Feedback loop

Act → Measure → Learn → Adjust

Applies to studying, product launches, and prompt engineering. No loop means no improvement - only hope.

LLM mental model

Tokenize → Attend → Predict next token

The model does not “know” facts - it predicts plausible text. Your job: supply context, verify outputs, define success.

RAG in one line

Retrieve relevant chunks → prepend to prompt → generate with citations

If retrieval fails, the answer will fail - fix search before rewriting prompts.

Quick checks before trusting AI output

• Can I verify this in a primary source?
• Would I stake a grade / job / budget on this without review?
• Did I give enough context in the prompt?

Go deeper

• Mental Models Hub
• How LLMs work
• Full courses