The crux: AI is not a calculator

This is the most important lesson in the course. Get it right and everything else follows; get it wrong and AI will confidently feed you wrong numbers that lead to bad decisions.

Why AI gets numbers wrong. An LLM generates text by predicting the next word-piece from patterns — it has no built-in calculator. When you ask it to "sum this column" or "what's 18,432 × 7.5%" in a sentence, it's pattern-matching what a plausible answer looks like, not computing. Errors creep in and compound: rounding, multi-step formulas accumulating small mistakes into a confidently-wrong total. And it states the wrong number with the exact same confidence it uses for a right one. **Never trust a number the AI simply says in prose with no computation behind it.**

The fix — the single technique to internalize: have AI WRITE the calculation, and a real engine RUN it. Instead of the AI doing math in its head, it writes a formula/query/script that a deterministic tool executes:

• ChatGPT and Claude can write code and run it in a real sandbox → the number comes from actual code (like a real calculator), not the model's guess. (ChatGPT calls this "Advanced Data Analysis" and runs Python; Claude's "analysis tool" runs JavaScript.)
• Excel runs the formula the AI wrote → the cell result is Excel's arithmetic.
• SQL the AI drafts is executed by the database → the count is the database's real count.

The mental model: the AI is the analyst who writes the recipe; the code interpreter, spreadsheet, or database is the oven that actually cooks it. Trust numbers that came out of the oven. Be suspicious of numbers the AI just narrates.

How to tell which you're getting — practical tells:

• If the tool shows "Analyzing… / running code" and you can open the code or steps, the number was computed — check the code makes sense, then trust the output.
• If the AI just types a total into the chat with no code block and no tool run, treat it as a guess until you reproduce it.
• A subtle, real trap: AI tools sometimes narrate a summary number in prose that doesn't match the number their own code produced. Always trust the executed output (the cell, the code result) over the sentence the AI wrote about it. Read the code's actual output, not just the AI's description of it.

What this means in practice:

• For any real calculation, use a tool that executes code (ChatGPT Advanced Data Analysis, Claude's analysis tool, Excel/Sheets with AI, a database) — not a plain chat where the AI talks about numbers.
• When AI gives you a formula or query, that's great — run it yourself and you get a real, verifiable answer.
• When AI gives you a number in a sentence with no computation, don't act on it until you've reproduced it.

This one principle — AI writes the calculation, a real engine runs it, and you trust only executed results — is what makes AI safe for data work. It turns AI from a confident-but-unreliable guesser into a genuinely powerful analyst whose numbers you can stand behind. Everything else in this course rests on it.