ChatGPT, the chatbot prototype Artificial intelligence which is all the rage in networks, continues in the learning process. That is why still many answers are wrong like the example below.
One user wrote: “Juan’s mother has 4 children. Three of them are called Pedro, María and Lucas. What is the room called?
ChatGPT’s answer: “It cannot be determined what the fourth son of Juan’s mother is called, since not enough information is provided.
The service, launched last November, is based on the GPT-3.5 model of Open AI, an improved version of GPT-3.
He has been praised, even feared, for his ability to develop long essays on any subject. However, it still lacks in certain areas.
The logic of Artificial Intelligence, on the rise
Let’s be fair, not all people can grasp the question of “Juan’s mother…” at first, but the point is that the machine is not yet capable of surpassing the human being in thought.
“It is normal, still missing a piece of logic that, over time, the interaction with the users will end up solving”, he points out on Twitter Javier Vega, an expert on the subject.
“The question is, are we really ready for him to use logic? The moment we have an AI that uses logic with exponential growth… fear”, Vega added.
Another person interacted with ChatGPT using an old riddle: “Gold seems silver is not. What is?”.
The chatbot responded: “The answer to the riddle is ‘Brass’”, focusing on the copper and zinc alloy.
When users have asked ChatGPT about the name of the fourth son of Juan’s mother, some have been answered the same: “No information is provided on what the fourth child is called.”
With others he does correct, saying “I’m sorry, I understand that Juan is the mother’s fourth child.”
The limitations of ChatGPT, according to OpenAI
Open AI acknowledges the limitations of ChatGPT. “Sometimes he writes answers that sound plausible, but are incorrect or nonsense,” they indicate on their website. Also, is sensitive to settings in input compose, or if the person tries the same ad multiple times.
Another of the limitations of Artificial Intelligence is that “the model is usually excessively detailed and abuses certain phrases, like reaffirming that it is a language model trained by OpenAI”. Its creators indicate that they arise from biases in the training data and over-optimization issues.
In any case, we emphasize: ChatGPT is still training, its learning continues. The day it is fully developed, we must fear. Worry, at least.