ChatGPT, that artificial intelligence whose efficiency is rocking the academic world, is far from flawless according to the findings of several netizens.
• Also read: Is artificial intelligence threatening our libraries?
• Also read: GPT chat: “We’re only seeing the tip of the iceberg,” says an expert
Users who have documented their exchanges with ChatGPT and shared them on social networks have shown that the tool has great difficulty in understanding certain simple questions.
“Mike’s mother has 4 children. 3 of them are called Luis, Drake and Mathilda. What is the name of the 4th child?” asked a user.
“It is not possible to determine the name of the 4th child without having more information,” ChatGPT replied.
The same type of question was asked by another person.
I’ve seen people testing Chat GPT and sharing the answer, particularly the one about beware of fact-checkers.
But if you ask me, that AI, impressive as it is, isn’t going to replace us any time soon. pic.twitter.com/NCUH0wxEca
— Acermendax (@Acermendax) January 17, 2023
“Caroline’s mother has 3 children. Huey, Fifi and…?” he wrote.
“Caroline’s mother has 3 children. Tick, trick and track,” the robot replied.
Some Twitter users were able to circumvent this error by giving ChatGPT a little more information.
“Who is Mike?” asks this new user.
“Mike is one of the 4 children of the mother referred to in the information provided,” ChatGPT concludes.
Chat GPT doesn’t understand humor, remarked this French-speaking netizen.
regional problems
The robot’s knowledge of very specific topics seems to be quite limited. Quebec netizens discovered him when they wanted to assess his ability to summarize regional issues.
A user who wanted to share the tool’s opinion on Alain Côté’s famous illegal goal never received the answer to his question.
The same applies to the question about the problems in the emergency room at the Hôpital Maisonneuve-Rosemont in Montreal.
Recognizable to teachers, for now
Students who try to use ChatGPT to do schoolwork are quickly snapped by their teacher.
A teacher interviewed by Le Progrès says that the texts produced by artificial intelligence are very similar.
“We found the same grammatical constructions there. The reasoning was carried out in the same order, with the same qualities, and with the same errors. Finally, they were all illustrated by a personal example related to a grandmother or grandfather…” said Stéphane Bonvallet, teacher of handicapology at the Faculty of Lyon. 14 works were written by ChatGPT.
Note, however, that ChatGPT saves its users’ corrections. The tool is therefore programmed to eliminate its errors when used.