AI tool limitations: what they cannot do¶
An overview to avoid bad surprises. The mechanisms behind these limits (knowledge cut-off, memory, varying answers) are covered in the tip What is an LLM?, and made-up answers in the tip Hallucinations: why AI makes things up. This page adds what matters for everyday use.
No access to your data by default¶
The assistant only knows the content of the conversation. It has no access to your files, your mailbox or your software, unless a feature of the tool explicitly allows it. If you want it to work on a text, you have to provide it, following the rules of the tip Data security and privacy.
Fragile calculations¶
A language model is not a calculator. It can get operations, totals or percentages wrong, especially on large amounts of data. Check numerical results, or do the calculation in a spreadsheet.
No accountability¶
The tool does not answer for what it produces and does not replace an expert's opinion on legal, medical or financial matters. The decision and final validation remain human.
In practice¶
| If you need… | Then… |
|---|---|
| Recent information | Provide the source or check it |
| A reliable calculation | Use a spreadsheet, or check the result |
| A specific document | Attach it instead of asking from memory |
| Content to validate | Don't rely on a single answer: compare or refine the instruction |
| A decision | Use AI as a thinking aid, not as a judge |