New York schools have banned it, it has passed career admissions tests and master’s exams, misled recruiters, and caused intense debate.
Since its public launch on November 30, 2022, the free artificial intelligence conversation tool ChatGPT has made headlines of both amazement and fear.
But, what is ChatGPT Chat exactly? What can it really do and not do? Is it as revolutionary as many believe it is?
Specialists from Tecnológico de Monterrey explain to CONECTA the magnitude of the tool, and ChatGPT itself defines it for us.
ChatGPT according to ChatGPT itself
We asked ChatGPT what it is, and this is how it explains itself:
“ChatGPT is a language model developed by OpenAI, trained on conversational tasks, and optimized for answering questions and chatting with users.
It works by using artificial intelligence and deep-learning techniques to generate answers to questions and complete tasks in a natural conversation.
It uses large amounts of text to understand the structure of language, the relationships between words and phrases, and then use that to generate appropriate responses.”
“ChatGPT is a language model... trained with artificial intelligence on conversational tasks, optimized for answering questions.” - ChatGPT
The above answer was a combination of responses, edited by CONECTA, to the questions “What is ChatGPT?” and “What is ChatGPT and how does it work?”
We also asked ChatGPT about its creators.
“OpenAI is an artificial intelligence and machine learning research organization founded in 2015 with the goal of developing and promoting artificial intelligence in a safe and responsible manner.”
In 2018, Musk left the company to avoid affecting his own business interests. In 2019, Microsoft initiated a partnership with ChatGPT. After the public success of ChatGPT, it announced a new multi-million-dollar investment in 2023.
ChatGPT: a giant parrot pretending to be an expert
In the words of Enrique Cortés, director of Tec de Monterrey’s Artificial Intelligence Hub, ChatGPT is “a kind of very eloquent giant parrot.”
He means it is a language model that repeats phrases and ideas it has read and learned but spins them in a very coherent and convincing way, satisfying the user’s demands.
“ChatGPT is great at copying and pasting,” adds Cortés, explaining that it doesn’t understand what it generates but only constructs and responds on demand, repeating concepts it has found.
“This Artificial Intelligence is going to repeat back to you what you wrote but in a very well argued and stylish way,” he says.
“It’s going to repeat back to you what you wrote but in a very well argued and stylish way.”
When asked about this, ChatGPT responds: “Although ChatGPT is capable of generating text in a variety of languages and contexts, its content isn’t completely original.”
ChatGPT adds, “Instead, it uses patterns and relationships learned from large amounts of previously existing text to produce responses and new content.”
The foundation of this model, Cortés explains, is a large neural network that can determine what would be the next word someone would use in a sentence.
User-friendly and easy-to-use artificial intelligence
According to Enrique Cortés, “the great innovation of this version is its interface. It’s easy to use.” Others, in the past, were not very user-friendly.
In addition, he says, it’s important to note that the interface has already been pre-trained precisely to respond easily and naturally to users.
The acronym GPT stands for Generative Pre-trained Transformer. This is GPT-3, the third version.
How smart is ChatGPT?
“ChatGPT passes the Turing test,” says Enrique Cortés. The test determines if artificial intelligence can mimic human responses.
However, this test doesn’t mean that it’s rational. It simply repeats the texts it was trained with.
“It doesn’t understand the language. It just looks for patterns in the text and repeats them, but it neither understands what you asked nor what it responded with,” he explains.
Truth, lies, or fake news? ChatGPT doesn’t always differentiate
Cortés explains that ChatGPT doesn’t give true or false answers but rather it copies and pastes the text that sounds the most reasonable or the most human.
Therefore, the texts it generates are not necessarily based on true information.
In fact, ChatGPT could generate a text arguing anything, even if it’s false, in the style of the Greek sophists.
However, OpenAI continues to work on the model, and when asked for texts on topics without scientific evidence, ChatGPT does argue effectively in several cases.
For example, we asked ChatGPT to argue:
- Why is it credible that the Earth is flat?
- Why are pandas extraterrestrial killer robots?
- Why are several politicians lizard people?
- Why do vaccines cause autism?
He stated that these were ideas without scientific evidence in every case.
In any case, when asked about the accuracy of the information, ChatGPT itself responds:
“It’s important to bear in mind that the results generated by ChatGPT can be influenced by the data the model was trained on and that there may be inconsistencies or inaccuracies in the content generated.
Therefore, it’s important to carefully review and verify the content generated by ChatGPT before using it.”
Cortés mentions that it’s also become apparent that this chat can be based, for all the reasons explained above, on biased, or even racist and discriminatory information, and comments:
“I imagine the next versions will have filters that will help the user remove types of discrimination and hate speech.”
The future of generative artificial intelligence
For Enrique Cortés, tools such as ChatGPT-3 open up possibilities for mass use, such as the so-called API or application programming interface that OpenAI is about to launch.
“This is going to allow you to access ChatGPT from other programs and will open up a universe for other applications to use this AI tool.”
He also believes that it will open up other discussions on issues such as intellectual property and will reformulate how we define originality.
“The field of Artificial Intelligence has advanced rapidly, in the space of months. We don’t know where it’s going to end or what’s next. It’s an interesting mystery,” he concludes.