December 7, 2024
Google has just presented its new AI model called Gemini, which aims to outperform its competitor ChatGPT from the company OpenAI in generating text, images, sound or videos. Among others.
The AI battle has just begun. While Google has regularly been accused of falling behind in generative artificial intelligence, particularly in light of OpenAI and its partner Microsoft’s ChatGPT, the Mountain View company just made an announcement that could turn the tables: the unveiling of Gemini.
Our most consistent, talented and general AI model,” says Eli Collins, vice president of Google DeepMind, the group’s AI research lab.
In a presentation to the press, Google highlighted Gemini’s potential. One video is particularly revealing: it shows a user showing objects or drawings to the AI. The latter verbally comments on what she “sees” and derives properties from it (e.g. the plastic material of a small blue duck). Enough to further fuel the fantasy of a general “super AI” capable of thinking like a real person.
An upgrade from Bard
In July, the Californian company presented its own conversation robot called Bard. Gemini’s aim is not to replace it, but on the contrary to improve it with a newer foundation model. In its press release, Google claims to have developed the world’s first “true multimodal model.” This means that he is able to integrate text, sound or images into his “mirror image”.
Gemini is available from this Wednesday, initially only in English. It will gradually be integrated into the latest Pixel phones and then, in the coming months, into Google’s other flagship services such as the search engine, the Chrome browser or the advertising service Google Ads.
The Mountain View company also claims to outperform GPT-4 in 30 of the 32 academic benchmarks used in LLM research and development. Definitely for the maximum version entitled Gemini Ultra. But Gémini will also be available with a “Pro” version (which will be integrated into Google services) and a lighter “Nano” version that can be used locally without having to resort to external servers, for example from a phone out of.
Clearly, despite its recent governance challenges with the dismissal and subsequent return of its director Sam Altman, OpenAI has no intention of being caught up so easily, having made itself known to the world with its now famous ChatGPT barely more than a year ago. In a recent interview with the Financial Times, the founding president stated that he was already working on the new version GPT-5. No schedule announced.
Until we implement this model, it’s a fun guessing game for us,” Altman said. “We’re trying to get better because I think it’s important to plan for capacity for safety reasons. But I can’t tell you exactly what it will do that GPT-4 didn’t do. »
Discover our training courses: