After several months of work and research, Google officially announced the deployment of Gemini, its new AI. With this tool, the American giant hopes to dethrone OpenAI, the creator of ChatGPT. Bet won?
In recent months, Microsoft has increased the integration of its artificial intelligence (AI) into the vast majority of its products and services. Thanks to its partnership with Open AI, the company was able to get ahead of the competition. However, Google may well have restarted the race for artificial intelligence. This Wednesday, December 6, 2023, the Mountain View firm announced that it was starting to deploy Gemini, its brand new AI tool. “Gemini is the model of artificial intelligence [IA] the most powerful and generalist that we have published”explains Sundar Pichai, CEO of Google in its official press release published on the company blog. Before adding: “This new era of models represents one of the greatest scientific and engineering efforts we have undertaken since the founding of Google.”
Google Gemini: an AI that intends to surpass ChatGPT
As confirmed by a test video broadcast by Google during a press presentation, Gemini is a so-called “multimodal” model. In other words, it is an artificial intelligence capable of analyzing sources of information that are very different from each other. Indeed, Gemini is able to both “to understand” a hand-written school exercise than deciphering videos or even lines of computer code. And that’s not all: the tool is also capable of recognizing the material of an analyzed object, of finding common points between several objects or even of understanding Chinese shadows.
According to Google’s evaluations, Gemini outperforms its direct competitor, GPT-4, in eight of the nine main tests recognized by the AI community. Among them, we find mathematics, translation and even Python programming. At the same time, the tool would also have solved 90% of the 57 tasks proposed during a test imagined by the University of Berkeley in which the machine was questioned on its ability to understand language models, law, history or more mathematics. For comparison, GPT-4 scored 87%, while its predecessor, GPT-3, scored 45%. “Gemini is the most versatile model we have developed to date. Its advanced capabilities will greatly benefit developers and businesses in building AI-driven applications.”indicates Demis Hassabis, director of Google DeepMind.
Google Gemini: an AI model available in three versions
To integrate Gemini into all its machines, Google plans to offer its model in three distinct versions, curiously described as “sizes”: Nano, Pro and Ultra.
- The Nano model is a lighter version of Gemini, specially designed for smartphones. Here, the objective is clear: that artificial intelligence can be used without an internet connection. Big news for application developers, who can already request access to Nano since Wednesday, December 6.
- The Gemini Pro model is a more versatile version, but also more advanced. This was notably designed to be integrated into all Google services and devices, including Bard, its chatbot, which is one of ChatGPT’s main competitors.
- Finally, the Gemini Ultra model, which was presented in a demonstration video, is the most efficient version of the LLM (large language models). In particular, it has been optimized to respond to particularly complex tasks. Unfortunately, this version is not yet ready for release. “We are currently carrying out in-depth verifications of the safety of the model, in particular using trusted third parties”explains Google.
For now, Gemini is aimed primarily at English-speaking users. Indeed, artificial intelligence responds much more precisely if the content is in English. For the moment, Google prefers to refine its model before releasing it in other languages. It seems that the Mountain View giant knows that it is expected to turn the corner on the quality of the products offered.