Google has just presented Gemini, its new generative artificial intelligence. This language model is presented as “the most effective and most general to date” by the company’s head, Sundar Pichai. A competitive alternative to ChatGPT, the OpenAI language model that was revealed a year ago and shook the industry.
But in its impressive demo video, it looks like Google fooled the Gemini’s abilities a bit. In any case, the company quietly claims that the video was edited for a more spectacular performance, Bloomberg noted.
“Reduced latency”
Concretely, this means that the demonstration should have been longer to respect actual response times. Therefore, it does not accurately reflect the capabilities of Gemini, which according to experts already surpasses ChatGPT 3.5. This is not anecdotal, since Gemini is made to interact in real time with a human. A post published on Google’s site dedicated to developers also details the making of the video. In particular, we learn that, for the demo, there has never been a real-time exchange and that the images are still.
Oriol Vinyals, project director at DeepMind (Google’s subsidiary dedicated to AI), explains on X (formerly Twitter) that the video was indeed shortened for “issues of brevity.” He specifies that the video “illustrates what a multimodal experience built with Gemini could be like” and that it aims to “inspire developers.”
Broadly speaking, Gemini is a multimodal artificial intelligence: it can manage several sources of information simultaneously. It accumulates general knowledge and problem solving, allowing it to accurately help those who use it. It’s already built into Google’s other AI, Bard, as well as the company’s smartphone, the Pixel 8 Pro, but not for Europe.
Source: BFM TV
