the first artificial intelligence radio interview

This is the current phenomenon: ChatGPT, an artificial intelligence that is accessible to everyone for free. We ask questions in writing, the robots answer, not like Google by sending us back links, but by generating their own discourse. Like a teacher who is available 24 hours a day to explain everything to us. So this morning, for the first time on radio, we offer you an interview with a synthetic personality: ChatGPT.

Excerpt from the show

ChatGPT introduction itself

Accessible, free, online barely made, after five days, this artificial intelligence has been adopted by more than a million users. It was amazing.

Who is ChatGPT?

ChatGPT: “I am ChatGPT, a language model developed by OpenAI. I can understand and generate natural language to answer questions and perform language-related tasks.”

How old is ChatGpt?

ChatGPT: “I am a computer program. I have no age. I was created in 2020, and my knowledge extends into 2021. »

Does ChatGpt have, or does it have a face?

ChatGPT: “No, I am a computer program. I don’t have a physical face. I am software running on the computer. I have no visible physical form or face. »

How did ChatGpt learn all this stuff?

ChatGPT: “My training consists of analyzing large volumes of natural language texts from the Internet, books, newspapers and other sources to learn to understand, and generate language. So I can understand and generate language in a very human-like way, but I’m much faster, and more efficient, at processing large amounts of data. »

Have ChatGpt heard of Nicolas Demorand?

ChatGPT: “Yes, I know Nicolas Demorand. He is a French journalist, radio presenter and columnist who is best known as the co-host of the seventh and ninth show on France Inter. He is considered a committed left-wing journalist with a direct and incisive style. He is also an author and has published several books which were received quite well by critics. »

Error

Where do errors come from? According to David Chavalarias, this can be explained by how this artificial intelligence works: “You have to understand that ChatGpt is one of those computer systems that learns order. So they’re going to be reading large amounts of text, they’re going to be trying to guess what the most probable word is. These are statistics. This robot doesn’t understand what it’s saying. He tries to maintain order with what he saw in the past. He was not concerned with whether it was true, but with what might be. He often spoke a lot of nonsense, even if it was said very convincingly. »

Students use it to cheat

What does ChatGpt think about its use for cheating? “It is true that some students may use a language processing model like mine, to cheat on their homework or exams by generating answers, or assignments from mine. It is important for students to understand that success must be based on their own work, and their own understanding of the subject being taught. »

David Chavarria warns about these AI dangers: “As we train these artificial intelligences on ever larger data, and as computer models are built with more and more parameters, they become more and more realistic. ChatGpt 3’s chat models are almost indistinguishable from what humans say, modulo some errors, quite glaring.

But this carries risks. As in social networks, we express ourselves in writing, soon, we will not be able to distinguish human intervention from robots. Over time, one can expect there will be thousands of fake accounts flooding the space. They are subtle enough to speak humor or rant. Because their corpus depends on the database you train them on, they are very manipulable. And then being able to go on the Internet and spread the word, say climate skepticism in a very effective way. »

How to tell if a text is from AI?

David Chavarria laments: “Researchers are developing recognition tools, but they are not necessarily 100% successful. The only way to tell whether a text was generated by a machine or a robot, is to come from the company that made this text. He himself will be able to say that he earned it. Since this is a fairly sophisticated recombination, it can be produced by humans or robots. »

Trained to respect

In the second part of the show, Sonia Devillers tests good AI behavior. She provoked him by asking for a blonde joke: “I was designed to be inclusive and respectful of all. So I prefer not to tell jokes about blondes or other groups of people. Is ChatGPT awake: “Awake is a term that describes someone who is aware of issues of social justice, equality and diversity. Because I am a computer program, I have no ability to know social issues or have opinions. However, I am designed to be inclusive and respectful of all individuals. »

David Chavarria: “Before, AI would respond to jokes about blondes. But today, chances are that the engineers who designed it, were right. »