- Google calls it “the new era of Gemini” because they are convinced that it will change the world.
- This new multimodal artificial intelligence works with text, audio, video, images, code, and other content.
Forget what you know or have tried so far about Google AI. Gemini is the artificial intelligence that Google has been developing for a decade. The AI that was going to give it a monopoly on the market, like Google’s search engine did three decades ago… if it were not due to the overwhelming emergence of OpenAI and GPT, which caught everyone by surprise.
Google has always boasted of being the pioneer company in the artificial intelligence market . He has been developing AI models for almost a decade. Surely Gemini was going to be a great debut, but OpenAI beat them to it with ChatGPT.
Today, according to Google CEO Sundar Pichai, a new era begins: the era of Gemini, multimodal artificial intelligence capable of working with text, audio, video, images, code, and other content. And it’s now available within Bard.
What is Gemini, the new AI that surpasses GPT-4
Companies like OpenAI, Meta and others have specialized language models for each task. For example, ChatGPT is text-based, DALL-E is image-based, and Whisper is audio-based. Each AI specializes in a “sense.”
If ChatGPT has a request for images, it uses DALL-E to generate it. They collaborate with each other. Gemini’s approach is very different.
Gemini is Google’s new large language model (LLM) that, according to its creators, far outperforms GPT-4. It is more powerful in 30 of 32 benchmarks where they have tested it.
Gemini is a multimodal AI : it can work at the same time with text, video, images, audio, code, etc. This, in theory, will make it faster and more accurate than using several specialized AIs. Google assures that, over time, it will add new “senses”. Including the touch of robots.
Gemini also promises to be the best at correcting and reviewing code. It integrates the new AlphaCode 2 version, which is twice as powerful as its predecessor. And, according to Google, it outperforms 85% of the competition, generating secure and stable code.
The three versions of Gemini
Another interesting feature is that there are three versions of the AI: Gemini Nano, Gemini Pro, and Gemini Ultra.
Gemini Nano is a shortened text-based version, like ChatGPT, that can work on mobile phones in local mode, without needing to access the cloud. It has just been integrated into the Pixel 8 Pro.
Gemini Pro will be the version that will be integrated into Google services, with functions similar to ChatGPT, DALL-E, Whisper, Stable Diffusion, etc. It will be the one used by the majority of individual users on computers, or on mobile phones connected to the cloud. Gemini Pro has just been integrated into Google Bard.
Finally, Gemini Ultra, which won’t be available until the beginning of the year, is the premium version for servers and enterprises.
In its blog, Google is convinced that Gemini is going to mark a before and after in the artificial intelligence market, as it did with its search engine. The theory is very promising, but how it works in practice remains to be seen. As we have mentioned, Gemini Pro is now available through Google Bard, although only in English, and for Anglo-Saxon countries.
