In recent days, half the world has been revolutionized by the trendy artificial intelligence ChatGPT, an AI capable of answering all kinds of questions with answers that are generally quite accurate. From history to programming, from gastronomy to health, ChatGPT amazes with every answer and it is no wonder that many practical uses of this artificial intelligence feature have already begun to be formulated.

I specify that this article is not a criticism of the whole. On the contrary, I proceed from sincere and unequivocal praise of what has been achieved. In addition, I have personally verified that in matters where caution is required, such as health, ChatGPT always shows relative answers and directs the user to a doctor instead of giving advice that could be dangerous in certain circumstances.

In the past, we have already seen some AIs begin to reproduce biases caused by their social interactions, biases in the resources used for their learning, or a combination of both. The creators of ChatGPT seem to have taken these kinds of issues into account and thus their AI seems to have been properly trained to avoid such issues.

Having made these clarifications and restated my opinion, more than positive, about ChatGPT, it is no less true that I see some opinions that confirm that this tool can already be used as a source of universal information, which seems to be insufficient. me because it is dangerous because their answers are not always correct and what is worse, the system does not show us this when it shows real information and when what it tells us is the result of its “imagination”.

So if we start from the false premise that everything ChatGPT tells us is true and assume that its answers are always reliable, sooner or later we will find that, even if it is without bad intentions, this artificial intelligence has “slipped” . into us”. ». Here are three errors caused by different causes.

Error of ignorance

The most logical and understandable of all. ChatGPT doesn’t know everything, even if it tries, and the information it has can become out of date over time… and may even be from the moment it is launched. Want a pretty clear example? This is what happens when you ask for the latest version of Microsoft’s operating system:

Yes, as AI claims; Windows 11 does not exist, I wonder what operating system we have been reporting on MuyComputer for more than a year.

Error due to incorrect resources

Valencian paella, the real Valencian paella, has a very specific recipe. And no, this is not a criticism of other excellent rice dishes, cooked in a similar way, which the biggest purists are not afraid to label as “rice with stuff”. Now, like a true carbonara, it is not made with cream, to give an example, no, the Valencian paella does not contain chorizo, peas, chickpeas or in general many elements that we find in this recipe:

All kidding aside (I’m looking forward to seeing this ChatGPT joke on Twitter), the incorrect sourcing causes the information supplied by the AI ​​to be incorrect in this case.

Unreal content error

As I said before, I find this case the most disturbing and it has to do with one of his main strengths, creativity. You may remember that a few months ago we tested DALL-E 2, an artificial intelligence capable of creating images from text descriptions. ChatGPT also has creative potential, be it prose or verse.

The problem is that he uses it even when he shouldn’t. Something I was able to verify when I asked him about one of my favorite poems, Tristes Guerras, by Miguel Hernández. This is what he replied to me:

I admit that the first answer surprised me, because when he claims that “that wars are sad when they are not fought for love”, it comes close to the meaning of Hernández’s text. However, the article “Las” at the beginning of the title makes me suspicious. So I ask, and that’s when this somewhat childish-flavored piece of poetry appears, which in no way resembles a real poem:

sad wars
If society is not love.
Sad, sad.

sad weapons
if not words
Sad, sad.

sad men
unless they die of love
Sad, sad.

Miguel Hernández – Tristes Guerras (songbook and ballads of absence)

The problem isn’t that ChatGPT decided to explore its creative potential, that sounds great to me. The problem is that when consulting a real work, instead of reproducing or referencing it, the AI ​​will invent the text based on the opening sentence without warning the user.

ChatGPT is a masterpiece of artificial intelligence and I personally recommend you try it (registration is free). However, keep in mind that as you’ve seen here, he’s not infallible, so you should take his answers with minimal skepticism and confirm them with other sources before taking them for granted.