Updated on / dernière mise à jour : 12/19/2022
ChatGPT is currently available for free and we must admit that it is a revolutionary artificial intelligence conversational model.
The tool provides answers not only in English but also in several other languages including French.
To my question "Are ChatGPT answers to same questions always the same?", its provided answer (in French) was intelligible and precise but it did use an anglicism, the word "humaines-comme" which doesn't exist in French. Obviously, the AI made a direct – but erroneous – translation from the English word "humanlike" (comme in French means like in English).
So I took the opportunity to mention to ChatGPT that "humanlike in French is not humaine-comme but rather de type humain". Its answer was very interesting since the AI recognized the mistake. I then rewrote the same initial question to see if ChatGPT would use the same anglicism again or if it would rather opt for my own proposal ("de type humain"). Its answer was identical except for the anglicism which was indeed replaced by "de type humain"!
I also wondered why ChatGPT used this particular "humain-comme" concept? What was the origin of this mistake? I used the famous Google translator to realize that when we translate "humanlike" from English to French using this particular tool, the proposed word is also "Humain-Comme" (the same mistake).
In conclusion, ChatGPT is – in my opinion – a very interesting new tool, but many errors remain possible, both in terms of the answers provided and the overall quality in regard to language and grammar. Surprisingly, ChatGPT automatically corrected its error on the next query without any form of pre-implementation verification from a "human" specialist. This seems to open the door to possible biases on the part of the tool's user community since it seems users could directly mislead ChatGPT by including or proposing false solutions and/or information to the AI.
It is possible to try ChatGPT on this website: https://openai.com/blog/chatgpt/