The subsidiary of the Alphabet group had announced in disaster in early February the creation of Bard, overwhelmed by the arrival in November of ChatGPT, developed by the start-up OpenAI in collaboration with Microsoft.

Able to produce emails, essays or lines of code on request, ChatGPT is generating immense enthusiasm for generative artificial intelligence.

The use of Bard was initially limited to "trusted testers," before Tuesday's opening to the general public. However, the number of connections has been limited and a waiting list has been established to manage the request.

Access is currently only possible from the United States and the United Kingdom.

"As people start using Bard and testing its capabilities, they're going to surprise us," Google boss Sundar Pichai said in a message sent to staff and seen by AFP.

"Things are going to go wrong. But user feedback is key to improving the product and the underlying technology," he added.

The leader of the Californian group had been criticized internally on the hasty launch of Bard to catch up with Microsoft.

The interface consists of a website, separate from Google's search engine, with a space in which the user can type a question.

"Real world"

Asked by AFP what distinguishes it from ChatGPT, Bard replied that unlike its rival it was "able to access real-world information through Google's search engine".

The chatbot also pointed out that it is "still in development while ChatGPT is already available to the general public. This means that I am constantly learning and improving while ChatGPT will definitely remain unchanged."

"We learned a lot from testing Bard and the next major step in improving it is to receive feedback from more people," Google vice presidents Sissie Hsiao and Eli Collins wrote in a message posted on a Bard website.

"The more people who use it, the better the large language models (LLM, a program capable of generating answers to questions formulated in everyday language) to predict answers that are likely to be useful," the two officials explained.

By being fed with data and written conversations, LLM's algorithm can more accurately determine the relevant answer to a question.

Bard relies on LaMDA, a language model designed by Google to generate chatbots, the first version of which was unveiled by the Mountain View (California) group in 2021.

The two Google executives acknowledge that LLMs "are not without flaws", and can "confidently offer inaccurate, misleading or false information".

Google says it has put in place "safeguards" to contain the possibility of inaccurate or inappropriate responses, including limiting the length of exchanges in a dialogue between Bard and a user.

Since ChatGPT went online, several Internet users have sought to push the chatbot to its limits and generated absurd, even worrying responses.

In an interview with The New York Times, Sissie Hsiao and Eli Collins indicated that Google has not yet determined a business model and monetization strategy for Bard.

© 2023 AFP