Google Gemini AI it’s the new one artificial intelligence by Google. A tool that, already in the testing phase, seems to offer answers noticeably higher than the competition. And that it is potentially capable of revolutionize the user experience on the Internet.
Google Gemini AI is a multimodal model, with input understanding capabilities never seen before. Especially in case he is called to interpret a multimedia sourcesuch as a video.
Its real possibilities are yet to be discovered, but i the first test they are delivering truly impressive results. Both in terms of comparison with other AI and with the human being. Both in terms of possibilities and areas of use.
What is Google Gemini AI and how it works
Google Gemini AI is the new artificial intelligence tool generative pointing to beat ChatGPT’s competition. Indeed, according to Google, Gemini has already surpassed the GPT-4 developed by OpenAI.
The new artificial intelligence Google is capable of understand different types of input: from textuntil reaching ImagesThe audio eh video. Without forgetting the inclusion of code and a whole series of new “senses” which, according to internal sources, are under development.
A distinctive feature of Google Gemini AI is its multimodal nature from scratch. This means that the dataset with which the tool was trained is not limited to text, but also contains other media.
This assumption contributed to the creation of a much more fluid and complete understanding. Other generative AIs reason first and foremost textual terms. Even when they receive a video, audio or image as input.
Google Gemini AI is a multimodal AI in development, which has already outperformed OpenAI’s GPT-4.
The new artificial intelligence Google on the contrary he really manages to understand a multimedia source. And the promotional videos showing it in action are simply stunning.
It is possible to find gods online first experiments of Google Gemini AI in action. In one of these a developer asks the tool to describe actions captured by a camera. And artificial intelligence follows everything in real time, giving testimony that is decidedly in line with that of a human being.
Added to this are the first results obtained by subjecting Google Gemini AI to the so-called test MMLUdedicated to massive multitask language understanding. And even in this case the AI’s response goes beyond all expectations.
For the first time in history, in fact, a model he surpassed the human beingas well as the aforementioned GPT-4, in various activities related to comprehension, knowledge e problem solving.
Google artificial intelligence was tested on over 50 subjects: from history to medicine, from mathematics to ethics, passing through physics and law. Getting a final score of 89.8% of correct answers.
The potential of the new Google artificial intelligence
Google Gemini AI it is still under development. And one of the first objectives declared by its developers is integration with upcoming smartphones Google Pixel.
The idea is for Gemini to make the most of its narcotics understanding ability. In order to be able to help the user in the main ones daily activities.
But the possible applications of this generative AI go far beyond the sphere of the ordinary. In another demonstration video some scientists show Google Gemini generating new code.
A tool thanks to which the tool succeeds first and foremost read and interpret circa 200,000 scientific documents. The material is then cataloged and filtered based on various relevance criteria. Finally the above studies become the basis from which to start create new content: real forms of meta-knowledge.
Staying on the topic of code, Google’s new artificial intelligence is already able to “speak” to the main ones programming languages: Yes Python a Javapassing through C++ e Go.
Google Gemini AI speaks major programming languages and is capable of generating dynamically coding user interfaces.
And even in this case the first practical demonstrations show revolutionary possibilities of use. Google Gemini AI is in fact capable of developing sites and user interfaces that code themselves dynamically. This means pages and content that they change appearance in real timebased on the needs of the user who visits them.
In a test published by GoogleGemini accompanies a user in the creation of one landing page dedicated to a little girl’s birthday party.
The AI receives some generic instructions on interests of the birthday girl and immediately proposes gods template a tema. Without any type of code being entered.
But not only. Once some small doubts have been clarified, Gemini independently proposes gods themes around which to build the party real. Adding games, activity and even proposals for catering.
Towards Gemini Ultra and Gemini Pro
According to Google sources, Gemini AI will be released in three different model sizes. The first one should be called Gemini Nano and was created for direct installation on mobile devices such as smartphone e tablet.
The second is currently known as Gemini Pro and will directly challenge OpenAI’s GPT 3.5. In the end Gemini Ultrathe model currently being tested, which far surpassed the results of the GPT-4.
Gemini Nano is already available for smartphone Pixel 8 e Pixel 8 Pro. And Google has announced that it will be rolling out to others soon Android devices.
Gemini Pro can be tested for free online through Google Bardusing a account Google. Although, to date, the available version is highly limited.
In the end Gemini Ultra may be available to the public by 2024. But before the official release you will need to run controls in-depth: both in relation toalignment of the models, and to stem any cybersecurity risks.
To know more:
Artificial Intelligence: what it is and what it can do for us