ChatGPT is a conversational tool based on artificial intelligence (AI). Developed by OpenAI, a company specializing in AI, ChatGPT is able to exchange in the form of messages. It answers your questions, but also generates written content in response to a written command of specific text, called “prompt” in English. Since it works in several languages, we can communicate with it in French. So we came up with the idea of ​​letting ChatGPT speak for itself about what it is.

ChatGPT is an exciting advancement in language technology. It is a language model driven by OpenAI which is able to generate text on demand using state-of-the-art artificial intelligence. The name “ChatGPT” comes from the combination of the terms “chat” and “GPT”, which stand for “conversation” and “predictive language transduction model”, respectively. This reflects ChatGPT’s ability to convincingly simulate human conversations.

How does ChatGPT, this AI-based conversational agent, work?

ChatGPT was trained on a large body of text, including newspaper articles, novels, film and television scripts, and online conversations. This allows him to understand the context of a conversation and provide relevant and consistent answers.

One of the most important advantages of ChatGPT is that it is able to adapt to a wide range of topics and language styles. This means that it can be used for applications such as generating content for social networks, writing scripts for movies and TV series, or even creating chatbots for businesses.

Additionally, ChatGPT is constantly evolving through continuous learning. This means it can improve over time and become even better at simulating human conversations.

In summary, ChatGPT is an exceptional language model that offers many possibilities in the field of artificial intelligence. Its ability to understand context and generate compelling text makes it a valuable tool for businesses and content creators.

Can ChatGPT impersonate a human writer?

You tell us. Everything you just read was written by ChatGPT, in French, with a simple command: “ Write an introductory article about ChatGPT in a journalistic style (proof: the screenshot below). We simply did the layout and added a subtitle, a title and the chapter.

ChatGPT writes about ChatGPT // Source: Numerama screenshot

What do we think? Honestly, ChatGPT is amazing. We did not seek to specify our order or attempt several generations. We took the first text proposed by the artificial intelligence and copied it, in order to offer you this little experiment. But, at the same time, we are also satisfied with the written output of the machine. The text objectively describes what ChatGPT is, offers a translation of the name for our French readership and goes into the details of how it works without being too technical. In short, a text adapted for the general public.

A few remarks however. First, the style is somewhat laudatory. ChatGPT doesn’t hesitate to use ameliorative adjectives to talk about him and insists a bit too much on his exceptional abilities. On the other hand, we see that the software is based on summary logical articulations (“In addition”, “In summary”) which are not to be banned, but which would be used by a human with more moderation.

How to use OpenAI’s ChatGPT?

To start chatting with the AI, you just have to go on the OpenAI website then create an account. The site is in English, but it is quite easy to find your way around. Start by clicking on the “TRY ChatGPT” button, then let yourself be guided. An email address and a password are enough to create a profile.

The site is sometimes a victim of its own success. When too many people are connected, you are redirected to a page which always indicates in English, that the site is overloaded. You will then see the message “ChatGPT is at capacity right now”. You will then have to postpone your connection attempt.

Before the explosion of ChatGPT, OpenAI had already done a lot of work on GPT-3, its first successful tool for generating text from artificial intelligence. The American company has managed to transform its “simple” text editor into a real conversational tool, with which a human interacts with an AI.

How to know if a text was generated by ChatGPT

The availability of a tool such as ChatGPT raises questions about its use. Being able to effectively detect whether the AI ​​is at the origin of a written production then seems essential. Whether it’s spotting copies of students who may have tried to cheat or simply determining responsibility for leaked content, especially when you know that ChatGPT is lying with disconcerting aplomb.

To meet this need, OpenAI hastened to release at the end of January, a tool capable of detecting whether the text was generated by ChatGPT or not. However, it is not so easy to detect text from ChatGPT and the “AI Text Classifier” tool still lacks precision. It just says if the text is probably “likely”, or possibly “possibly” from an AI with a non-negligible margin of error. In 9% of cases, the tool attributes to the AI ​​a text written by the human, which could have consequences when we know that Science Po sanctions the use of ChatGPT.

All about ChatGPT

How does GPT work?

GPT, the artificial intelligence used in ChatGPT, is based on what is called the deep learning. It is an artificial reconstruction of a whole neural network on several layers, capable of learning. THE deep learning mimics the faculty of conceptualization specific to humans and which makes us able to distinguish a cat from a dog.

When a child shows a cat for the first time, asking what it is, if the answer is: “It’s a cat”, he sticks the picture of the cat to the word “cat”. When this child points to a dog and asks if it’s also a cat, they say “no, it’s a dog”, a new name/image association will be created. As the child is exposed to different cats and other feline-like animals, he will be able to distinguish any cat, regardless of size or color.

The AI ​​training works identically on a trial and error system. It is therefore necessary to present a fairly substantial amount of information so that the algorithm can be efficient and learn. To perform all these tasks, AI like ChatGPT requires significant computing power, provided by GPUs, to complete this learning phase. We also talk about training.

Who developed ChatGPT?

ChatGPT was developed by the OpenAI company. Co-founded in 2015 by Elon Musk and Sam Altman, the company specializes in the development of artificial intelligence. She is at the origin of DALL-E, an image generator created from AI and leads several projects such as Roboschool, a robotics simulation platform dedicated to developers.

OpenAI first developed an early version of GPT, Generative Pre-trained Transform, the language allowing ChatGPT to give its answers. The model then evolved to ChatGPT, which stands out from previous versions, and in particular from GPT-3, by its ability to exchange with a human. The tool not only reproduces a text, a poem, a song or a cooking recipe, it is also capable of conversing and answering questions put to it, creating a real dialogue. The GPT-4 version allows ChatGPT to be integrated into the Bing search engine. A major issue for Microsoft that risks disrupting the use of the Internet.

ChatGPT vs Bard, what’s the difference?

ChatGPT is developed by OpenAI and is based on the GPT language. Bard uses the LaMDA language model. This conversational artificial intelligence model is developed by Google, which intends to integrate it into its search engine.

Apart from their development company, the two AIs differ in the sources they were trained with.

OpenAI’s ChatGPT tool is unable to tell you who won the World Cup in 2022, nor inform you about the evolution of the war in Ukraine. The data with which ChatGPT was trained stops in 2021. Beyond that, the AI ​​has no data on which to base itself to provide accurate information. Nevertheless, “ChatGPT with Bing” overcomes this problem, because it will be connected to the Internet.

Bard, Google’s conversational neural agent, has access to the internet to deliver his knowledge. The recent announcement of Bard by the American giant does not allow us to say more. For the moment, only “trusted” testers have access to the Google tool.

Is ChatGPT available on iPhone?

ChatGPT won’t replace Siri anytime soon, especially because it doesn’t understand your context and won’t be able to give you the weather forecast or control your connected light bulbs. But you can have fun integrating ChatGPT into a Shortcut on iPhone.

ChatGPT speaks how many languages?

We asked the question to the interested party, who reminds us that the concept of language does not concern him as an artificial intelligence that composes text. See instead:

I’m an OpenAI-trained language model, so I don’t really speak languages. My goal is to generate text on demand using state-of-the-art artificial intelligence. I can be used to simulate human conversations in different styles and topics, but I am not able to speak multiple languages. I am a tool for generating text, not for speaking. »

Specifying our request, ChatGPT replied that it was done “ For generate text in english “. In short, it does not yet know that it can generate text in a language other than English, while generating text in a language other than English. On Twitter, many users show results in other languages (Spanish, German, Italian, Chinese, Japanese, Korean, Portuguese, Russian, etc.).

For further

The Ghostwriter machine writes by itself // Source: Arvind Sanjeev / Twitter


Subscribe to Numerama on Google News so you don’t miss any info!

Understand everything about experimenting with OpenAI, ChatGPT

California18

Welcome to California18, your number one source for Breaking News from the World. We’re dedicated to giving you the very best of News.

Leave a Reply