Google announces Gemini 1.5, the most powerful version of its AI model

Google today (15) presented: Gemini 1.5 is the next generation of the generative artificial intelligence (AI) model that powers the search giant’s chatbot. The latest version of the tool promises optimizations in processing capacity, reduced response time and greater assertiveness.

Built on Mixture of Experts (MoE) architecture, the update only runs a portion of the overall model during search, rather than the entire network. According to Google, this is Ability to process large amounts of information at the same time, faster and more efficiently.

Gemini’s successor, 1.5 model can handle up to 1 million coins simultaneouslyIt easily beats Gemini 1.0 Pro’s 32k tokens and OpenAI GPT-4’s 132k tokens. In the future, the developer plans to increase the number obtained in tests to 10 million tokens.

With the context window now announced, Gemini 1.5 can process information equivalent to 1 hour of video, 11 hours of audio, 30 thousand lines of code or more than 700 thousand words at a time.. In one demonstration, Google asked the AI ​​to analyze a PDF containing transcripts of the Apollo 11 mission, whose 402 pages were read in seconds.

Optimized performance and security tests

In the race to create the best productive AI on the market, Google emphasizes: The new generation Gemini outperforms its predecessor by 87% in benchmark tests, in a text, code, photo, audio and video review panel. The recently released Gemini Ultra has also been left behind, with the updated model proving to be 55% faster.

The technology giant also said that the new version is in the preparation phase. “extensive ethical and security testing”exposes the tool to a range of potential damages. The aim is to ensure that the engine is completely safe for a wider release.

Gemini 1.5 will initially be available to developers and enterprise users, before being released to the public on the chatbot of the same name, which has no actual date yet. Samsung is one of the companies that was able to access the solution in the testing phase through Vertex AI, which develops artificial intelligence capabilities for the Galaxy S24 series.