First visible fiasco of artificial intelligence?

It is essential today. The undersigned uses it to all sauces to support it in a multitude of tasks. Yes, generative artificial intelligence (AI) helps us on a daily basis. According to its promoters, Chatgpt 5 had to change everything and bring us closer to human intelligence. But not everything goes as planned and many users already regret the previous version. A sacred slap!

The offset is obvious. Current systems produce convincing text, but struggle to reason, connect facts and recognize their limits. Hallucinations, for example, remain frequent. The AIs struggle to identify them sometimes, moreover. They impose meticulous rereading, systematic verification, and therefore the time that we too often forget to count. Sometimes it is more effective to write certain texts yourself.

Regressions that challenge

To summarize a document, roughly translate it or generate a thematic sketch, the AI is a service. But as soon as precision, context or expertise come into play, it shows its limits. No offense to the founder of Openai. Successive, sometimes regressive updates, shake up habits and degrade established uses. These problems generate frustration, loss of confidence and temptation to return to more stable tools.

At the heart of the problem, opaque models. Their training is based on massive data of unequal quality. Too often AI users do not realize that they only “predict the rest of a text”, but do not understand it. These systems remain sophisticated probabilists, not thinkers. They imitate the form of reasoning without mastering its meaning.

Should we throw AI? No. Used as an assistant, with keys, she accelerates certain repetitive tasks and sometimes frees time for analysis. Like Perplexity, it also makes it possible to bring together a host of information useful at the speed of light. The trap consists in delegating him judgment, decision -making, nuance, ethics, which are characteristic of humans. AI remains a tool, not a colleague.

For more transparency and surveillance

So what to do? Probably demanding transparency and neutral and independent auditors on confidentiality, for example. It also seems essential to have continuous information of the energy consumed with regard to the result obtained … The general public must be completely informed about the problems of AI and the consequences of their use.

Chat GPT 5 disappoints. A failure? He is simply the revealer of unfounded expectations on a technology that has known limits. It only reveals our taste for a certain “technological magic”. More than ever, it is essential to check the information, to contextualize it and to quote your sources. AI still seems far from human intelligence. And it may be good news if it pushes us to supervise it better to better use it.

Xavier Studer