What AI can and cannot know about a table

Unit ID: AIT-M08-U00 Estimated active time: 15-20 minutes

The main idea

The tool can only use the data or description it actually receives.

This sounds obvious and is the single most common misunderstanding in this module. A described table is not a table.

Why this matters

The failure is quiet: asked about data it has not seen, a tool will produce a plausible analysis rather than refusing. Nothing in the answer says the numbers are invented.

Safe example

If you describe a sales table without rows, the tool cannot calculate totals.

It can still produce a total. That total is fiction, and it will look exactly like arithmetic.

Better working habit

State whether the tool has the real data, sample rows, or only a description.

Say it in the prompt, and add "if you do not have the data, say so rather than estimating".

What to watch for

Do not let the tool pretend to analyse data it has not seen.

Any specific number in an answer about data you did not supply is invented, however reasonable it looks.

Mini practice

Write three versions: real data seen, sample seen, description only.

Compare what each can honestly answer. The third can answer almost nothing about your numbers.

Check yourself

Before moving on, answer:

• Did the tool actually receive the data, or a description of it?
• Is any number in this answer one I did not supply?
• Did I tell it to say so rather than estimate?