SAN FRANCISCO, Dec 9 ― In its bid to impress, Google may have slightly exaggerated the possibilities of Gemini, its new multimodal language model. Designed to boost the performance of the artificial intelligence integrated into Google's various online tools, Gemini nevertheless heralds a new era.

According to Google, its brand-new multimodal large language model, Gemini, is a veritable revolution in the history of AI. It is theoretically capable of understanding, processing and analysing different types of media, from text to computer code, as well as audio, images and video. It will be available in three formats (Ultra, Pro and Nano), adapted to different types of task.

It promises unprecedented experiences, as Google has illustrated in a stunning video recently posted online. This shows how Gemini can answer a whole series of questions posed by a human, with impressive precision and consistency. The problem is that the video doesn't entirely reflect reality, reports Bloomberg. The questions put to Gemini are reported to be have been pre-selected, and the answers slightly modified to make them more to the point. “For the purposes of this demo, latency has been reduced and Gemini outputs have been shortened for brevity,” says Google in the YouTube description of its video demo. According to Bloomberg, the demonstration was not carried out by voice either, but via written prompts, which ultimately offers a much less fluid experience than the one seen on-screen.

Gemini was first announced last spring at the Google I/O developer conference. It is billed as a new generation of large-scale, pre-trained multimodal language model, capable of understanding natural human language, deciphering text whatever the language used, but also of processing information from an image, audio or video file. The idea is for it to feed into all Google's online services, from Bard to Gmail to Assistant, to boost their processing capabilities

Google recently integrated Gemini Pro into the AI chatbot Bard. This update is available in 170 countries, but only in English. Over the coming months, Google promises to integrate Gemini Pro into Bard in other languages and territories. ― ETX Studio