Gemini, Google’s top AI model: learn all the details

This is what everyone was waiting for, the announcement Google DeepMind with twina model of artificial intelligence Multimodal and flexible (artificial intelligence) promises to revolutionize the way technology is intertwined with everyday life and business development.

Sundar Pichai, CEO Google And the alphabetHe stressed that this new model, which is part of the company’s initiative to be “AI first,” has cutting-edge performance on several academic benchmarks and is capable of outperforming human experts on multi-concept language comprehension tests.

twinwhich is optimized to run on devices with different capabilities (Ultra, Pro, and Nano), demonstrated superior performance on 30 of 32 academic tests applied to language models, including a score of 90.0% on Multi-Concept Large-Scale Language Understanding (MMLU), beating The analytical ability of human specialists in fields such as mathematics, physics, history, and medicine.

Sundar Pichai emphasized the continued acceleration of AI progress in the enterprise and the promise that these advances will be of global benefit.

The new system, which is the result of an extensive collaborative effort in Googleis designed to efficiently understand and analyze information, covering text, code, audio, image and video.

Unlike previous models that combined separate AI components, twin Since its inception, it has been trained to manage multiple modalities in an integrated manner. This allowed him to improve his understanding and thinking.

Therefore, this technology represents one of the largest efforts in the field of science and engineering developed by Google in its modern history.

Gemini 1.0 It contains advanced capabilities for complex multi-modal reasoning, which helps discern hidden insights in large amounts of data and enhances AI’s role as an intuitive assistant that can help in fields as diverse as science, finance, and programming.

High performance in programming tasks and the ability to understand, explain and generate high-quality programming codes in the most widely used programming languages. Gemini Ultra As one of the basic models in the world of programming.

Specialized version, Alpha code 2showed significant improvements over its predecessor, resulting in increased effectiveness in solving complex programming problems.

Google claims to have conducted the most comprehensive security assessments regarding Bias and toxicity because twin It is integrated into Google products such as the digital assistant A poet And the Pixel 8 Prowhile Gemini Pro It will be accessible to developers and enterprise customers through Google Artificial Intelligence Studio And Google Cloud Vertex AI From December 13, 2023.

In its development, Google focused on accountability and security, working alongside internal and external experts to identify and mitigate potential risks.

twin It includes powerful security classifiers and filters to handle sensitive content, and has leveraged tools such as True toxicity claims Subordinate Allen Institute for Artificial Intelligence For evaluation during typical training stages.

The company also highlighted its cooperation with organizations such as MLCommons, Frontier Model Forum And the Secure Artificial Intelligence Framework (SAIF)With the aim of establishing safe practices in the field of artificial intelligence.

Bard will use an improved version of Gemini Pro It represents the biggest update since launch, although it is currently in English.

he Pixel 8 Pro It is the first smartphone designed for use Gemini Nano Which already improves the existing features of the Recorder app and is expected to be included in other messaging apps like WhatsApp.

In the coming months, it will be released twin And in other services such as Search, Ads, Chrome And Amnesty International Duet. in addition to, twin It is tested in seekwhich has already reduced response time by 40% in English for users in the United States.

Starting December 13, 2023, developers and enterprise customers will have access to Gemini Pro across the Gemini API.

Google Artificial Intelligence Studio Provides tools for rapid prototyping and application release, and Vertex Artificial Intelligence Offers a managed AI platform with additional security controls. In addition, Android developers will be able to work with them Gemini Nano during AICore in Android 14initially available on devices Pixel 8 Pro.

Gemini Ultra It is in the final stage of trust and security checks and will be trialled to a select group before its general release early next year. Likewise, Google plans to launch cool advanced with Gemini Ultra To provide advanced artificial intelligence experience.