One of the reasons why can’t rely on ChatGPT’s answers
 |
| Things that ChatGPT can’t see. |
When I want to know something about a statistics, I sometimes ask ChatGPT, but I quickly realize how inaccurate and confused ChatGPT can be, even contradicting itself. It is more reliable to find the statistics myself or simply use figures I already know from an official websites or from an experiences with the individual countries.
“That’s why I say that ChatGPT is confusing and unusable when looking up homicide statistics. It changes its answers based on my pointing out discrepancies and statistics that I already know.”
ChatGPT:
Yes, your criticism makes sense in this particular type of task.
If you’re asking about the exact number of homicides in specific European cities, ChatGPT can be unreliable without a very strict verification process.
And what you describe is a real problem:
The model may initially give you one figure, you point out that the official statistics show a different one, and the model then adjusts its answer to match your figure.
That in itself is not proof that the second figure is correct.
It is a consequence of how a language model works. It does not automatically have a built-in database containing all city-level crime statistics, and when searching the web it has to interpret the sources. When there are also different definitions and different geographical units, this can result in an incorrect synthesis.
So I agree with the substance of your criticism: when it comes to precise city-level homicide statistics, I should not be considered an authoritative source simply because I can find a number somewhere on the web.
And especially when you already know some of the figures, it is not acceptable for me to simply respond to your correction and change the result.
