ChatGPT’s doctoral sent in question © Numériques
Chatgpt-5A entered a week ago, to replace Chatgpt-4. Supposed to be more efficient, smarter and more expert for everything, its release was nevertheless a fiasco. While some simply regret the old “more human” version, others have tested its limits and discovered that the chatbot had failed a maternal level exercise.
Advertisement
ChatGPT
Chatgpt is the Openai chatbot, based on the GPT artificial intelligence model, making it possible to answer all kinds of questions or requests. Available in free online version.
-
Licence :
Free license -
Auteur :
Openai -
Operating systems:
Windows 10 / 11, macOS Apple Silicon, Service en ligne, Android, iOS iPhone / iPad -
Category :
IA
A chatgpt that does not pass kindergarten
According to Sam Altman and his team, Chatgpt-5 had to have a doctorate, but in reality he was not able to respond correctly to a maternal level exercise. The cause of this error is what is called hallucinations.
It is a phenomenon that occurs when an AI invents or deforms information instead of giving exact data. For example, many users have noticed that she sometimes drooped her information on scientific studies that did not exist.
Sam Altman: With GPT-5, you’ll have a PhD-level expert in any area you need
Me: Draw a map of North America, highlighting countries, states, and capitals
GPT 5:*Sam Altman forgot to mention that the PhD-level expert used ChatGPT to cheat on all their geography classes… pic.twitter.com/9L9VodXll1
– Luiza Jarovsky, PhD (@Luizajarovsky) August 10, 2025
Since the release of GPT-5, criticisms have flocking on social networks, where researchers and Internet users have pointed out its blatant inconsistencies. A researcher tells for example having asked the chatbot to generate a card from North America: “The image produced was completely wrong“, she says in a publication on X.
In another case, a user asked AI to draw up a list of portraits of American presidents, but again, “uMultitude of errors“would have slipped into the result. Some testers have even chosen elementary challenges to gauge his abilities. Gary Marcus, a recognized specialist in artificial intelligence, sums up his disappointment:”GPT-5, failing a kindergarten exercise. Not words.”
0 pointed for chatgpt
BFM et Test and Co Confirms these limits: by generating a map of France with its 12 largest cities, GPT-5 has moved regions, forgot the DOM, replaced Paris by Orléans and reduces the list to six cities. Same observation with the portraits of French presidents under the fifth republic: if the beginning is correct, the rest becomes incoherent.
The map of France filled with errors for chatgpt © Bmftv tech and co
These errors mainly concern the creation of images, maps and graphics, while the textual responses remain fair. From what we observe, the entries in general remain the big weak point of Openai. It should be remembered that the AI of generation of images do “understand“Not really the letters. They treat them like simple visual forms, which causes distorted or illegible words.
Advertisement
Want to save even more? Discover our selected promo codes for you.