Google has announced the launch of the first version (1.0) of Gemini, a multimodal and flexible Artificial Intelligence (AI) that comes in three levels of use - Gemini Ultra, Gemini Pro and Gemini Nano - for different applications, such as devices or data centers.
At the end of last March, 'The Information' announced that Google's artificial intelligence (AI) team and DeepMind were working on a new initiative with which they sought to compete and overcome the dominance of the ChatGPT developer.
This initiative, which was known internally as Gemini, brought together both teams dedicated to AI at Alphabet, Google's parent company, to create a new AI model, of which no data was disclosed.
Google has now presented the first version of Gemini (1.0), an AI model that begins a "new era of models" and that "represents one of Google's greatest science and engineering efforts," according to the CEO of the company. signature, Sundar Pichai, in a statement.
Google DeepMind CEO and co-founder Demis Hassabis said: "Gemini has been built from the ground up to be natively multimodal, meaning it can seamlessly understand, operate, and combine different types of information," including text, code, audio, image and video."
This AI is also characterized by being flexible, so that it can be run efficiently both in data centers and on mobile devices, which is why it has been optimized in three different sizes.
Gemini Ultra, for its part, is the largest and most capable model to carry out highly complex tasks. Google has noted, on the other hand, that the Gemini Pro is ideal for scaling across a wide range of tasks and that the Gemini Nano is the most efficient for on-device tasks.
Google has explained that Gemini Ultra's performance exceeds current state-of-the-art results in 30 of the 32 academic benchmarks used in large language model (LLM) research and development.
This version of Google's new AI also outperforms human experts in Massive Multitasking Language Understanding (MMLU) by 90 percent, using a combination of 57 subjects, such as mathematics, physics, history, law or medicine.
On the other hand, it has indicated that with the image benchmarks it has tested for its development, Gemini Ultra "Outperformed previous state-of-the-art models" without the help of optical character recognition (OCR) systems, which extract text for further processing.
This AI can also extract information "from hundreds of thousands of documents" by reading, filtering and understanding information, which the company says will help achieve new advances at digital speeds in many fields, from science to finance. ".
At the moment, this first version of Gemini can understand, explain, and generate high-quality code in the most popular programming languages, such as Python, Java C, and Go. Likewise, it can be used as an engine for coding systems such as AlphaCode 2, which excels in solving programming problems that go beyond coding and involve complex mathematics and theoretical computer science.
Gemini 1.0 is being implemented in different Google products and platforms, including Bard, which will now use an improved version of Gemini pro for more advanced reasoning, planning and understanding.
Likewise, this technology will come to Pixel - more specifically, to Pixel 8 Pro, with the Gemini Nano version - to boost functions such as 'Summarize' in the Recorder application and the implementation of 'Smart Reply' in Gboard.
In the coming months, Gemini will be available in Google services such as Search, Ads, Chrome and Duet AI and is also being tested to make the Generative Search Experience (SGE) faster for users.
Starting December 13, developers will also be able to access Gemini Pro through the Gemini API in Goole AI Studio or Google Cloud Vertex AI.
Regarding Gemini Ultra, the American firm has explained that it is currently completing "exhaustive trust and security checks", which includes the training of external teams, as well as refining the model based on the so-called Reinforcement Learning from Human Feedback ( RLHF).
