The new application is available on iPhones in the United States, for starters, and should arrive "soon" in other countries and on mobile phones operated by Android (Google), according to a statement from the Californian start-up.

Free, it allows, like the website, to discuss with the chatbot and especially to ask it to write messages, explain technical concepts, suggest ideas, summarize notes ...

OpenAI, for example, promises to "get accurate information without having to sort between ads or multiple results," the current search engine model. But on first opening, the app also warns early that ChatGPT can "provide inaccurate information about people, places, or facts."

Launched at the end of November, ChatGPT's website surpassed one million users in one week, a record. Two months later, the service already had some 100 million monthly active users, another record according to a UBS study relayed by the press.

Microsoft, the main investor in OpenAI, has integrated the broad language model on which ChatGPT is based into Bing, its search engine, and Google is about to launch a test version with generative AI.

This ultra-fast adoption of ChatGPT and other generative AI software (computer code, images, sound, video) is causing fundamental concerns across many industries.

Teachers are seeing their students delegate their essays to ChatGPT, many administrative and creative jobs are at risk, politicians fear that this technology will foster increasingly sophisticated misinformation, and lawsuits have been launched over intellectual property issues.

Sam Altman, the boss of OpenAI, advocates government intervention to better regulate artificial intelligence. At a Senate hearing Tuesday, he said AI has "the potential to improve just about every aspect of our lives," but also "creates serious risks."

How ChatGPT is trained to answer © us Valentin RAKOVSKY / AFP

OpenAI wants to achieve so-called "general" AI, i.e. programs with human cognitive abilities.

The success of ChatGPT also creates opportunities for hackers: Meta (Facebook, Instagram) warned earlier this month against fake programs masquerading as AI tools.

© 2023 AFP