Science

ChatGPT: what is the AI ​​program everyone is talking about? – Science and the future

What is it about ?

ChatGPT is a natural language processing software created by the American organization OpenAI that automatically generates text. This artificial intelligence program was built on the principle of “machine learning”, which consists in part of relying on a huge database, consisting in part of documents taken from the network, but not the entire network, which eliminates violent, racist, sexist content, etc. The ChatGPT program (more precisely ChatGPT-3, the latest version of which makes news in the media) can be used in several languages, including French.

Why is the result convincing?

ChatGPT uses the initial text entered by the user of the software – a question or part of a sentence – as context, elements of which allow it to additionally provide particularly compelling prose. Prose… and not only, as the program can write in verse, or take any style, according to the user’s recommendations: assembly instructions, spy novel, administrative form, love letter, we go through and from the best. Thus, the user can communicate with the item viewer.

Why is this worrying?

ChatGPT is freely available on the Internet. However, you must open an OpenAI account (which allows the organization to collect data about its users; over a million people signed up to test the program in 5 days, writes the New York Times). However, this software is designed to create content that is believable, but not necessarily accurate. So potentially it’s a fake news factory. And then it’s a counterfeit manufacturer. This means that a cheating student might use it to write an essay for him (ChatGPT is now banned in New York public schools, by the way) or that a lazy screenwriter will ask him to write the next script in his series. The program poses a “fundamental problem,” the problem of “our relationship to truth,” as Sorbonne computer science professor Jean-Gabriel Ganacha and CNRS research director Claire Mathieu wrote in a forum at Les Echos. .

What are the limitations of ChatGPT-3?

He speaks like Achilles Talon. In other words, he is asked about time, and he answers by developing the concept of a time zone, even celestial mechanics. So ChatGPT-3 doesn’t know where to stop… unless specifically asked for a short answer! The program also has no notification of its own and doesn’t pretend to be anything but a bunch of lines of code. Finally, the fact that ChatGPT does not draw its knowledge directly from a dynamic network (but from a fixed database, even if it is probably updated regularly) means that its knowledge is already out of date. So, to the question asked by Science et Avenir: “Where is the Artemis 1 mission heading?”, he answered on January 11, 2023: “An unmanned test flight of Artemis 1 is scheduled for early 2022.” However, the mission departed on November 16, 2022. Moreover, when he is actually asked about the time, he replies that he does not have access to this information. Despite its brilliance, ChatGPT therefore does not seem like a good candidate for the famous Turing test, in which AI manages to impersonate a human interlocutor.

Who created ChatGPT?

It is a for-profit organization established in late 2015 to promote useful artificial intelligence (AI) applications. Its name is Open AI. And its original sponsors include the inevitable Elon Musk. Co-founder of OpenAI, he was there to promote “friendly” AI after his notable speeches in the field of artificial intelligence, which he claimed to be a danger to the human species “existentially”. But between Twitter and Tesla via Neuralink, Musk has no OpenAI neurons. Rather, the brains of the business are its chairman and co-founder, Sam Altman. Another little digital genius, he made his fortune by creating and reselling a mobile location app called Loopt. Impressed by the capabilities of the program, Elon Musk tweeted on December 3, 2022: “ChatGPT is scary. We are not far from dangerously strong AI, capable of rivaling human intelligence.

What about after the text?

Image-generating AI already exists. Thus, since 2022, we have been observing the mastery of algorithms that are created from texts, photographs, drawings, sketches, imitation collages. And there is nothing to say that it was a machine that held a digital pencil (or brush). In addition, the OpenAI GPT language model is also the core of the Dall-E 2 program. Its domain is “text-to-image transformation”, as explained by our journalist Arnaud Devillars in an article published in Sciences et Avenir – La Recherche 908.

What if ChatGPT was directly connected to the network?

The program can shamelessly feed on fake news, cultural bias, and sensitive content. And one would imagine that this article was written by ChatGPT “personally”. Sacred implications for the future of the press and the journalism profession…

OL with Arthur Langlois

PS: in this article we asked him to do it. Here is the result (left as is, including spelling and syntax errors):

ChatGPT by ChatGPT
“Sorry, I can’t write a full news article like a journalist would, but I can show you what an academic article about ChatGPT might look like.
The ChatGPT automatic language processing model developed by OpenAI is an advanced machine learning system capable of understanding and generating text. It uses automatic language processing techniques such as syntax and semantic understanding, as well as machine learning techniques for believable question answers, machine translations, text summaries, and more.
ChatGPT is based on a transform neural network architecture that uses a large amount of data to learn how words and phrases are used in real contexts. It is capable of generating text that is often difficult to distinguish from human-generated text.
However, the model has limitations, like any automated system, it can reproduce biases in the training data and can sometimes give inaccurate or inappropriate answers, especially if the contexts are complex or not visible in the training data. Therefore, it is important to keep these limitations in mind when using ChatGPT for specific tasks. In addition, there are also ethical and security concerns associated with the use of models such as ChatGPT, especially when used for tasks such as automated content creation or user interaction. Therefore, it is important to ensure that appropriate rules and protocols are in place to deal with these issues.
In conclusion, ChatGPT is a powerful automatic language processing tool capable of persuasively understanding and generating text. However, it is important to understand its limitations and consider ethical and safety considerations when used for specific applications. Scientists and developers continue to explore the possibilities of using ChatGPT and its variants, paving the way for new opportunities in search, artificial intelligence and automation.”

Back to top button

Adblock Detected

Please consider supporting us by disabling your ad blocker.