Google introduces Gemini 1.5, the next generation of the company's artificial intelligence model – Google

Google started Power point Gemini 1.5, the new version of the core AI model the company is developing. Gemini 1.5 is a completely renewed version, with improvements that allow more efficient training and exploitation of the model. Its first available version, Gemini 1.5 Pro, combines the capabilities of the larger model, Ultra 1.0, into a cutting-edge package that requires fewer computing resources.

Pro processes up to 1 million objects, and ingests massive amounts of data, such as hours-long videos or 30,000-line code bases. Early adopters who participate in the beta run of the model will have the opportunity to enjoy its full potential for free. Despite being a 'lighter' model, the Pro proved to be the Ultra's equal at a variety of tasks. In a presentation hosted by Google, the Pro was shown to be able to accurately describe details in a silent movie and translate from English to the endangered Kalamang language, simply by incorporating reference texts.

The new architecture adopts a “mixed-major” design, dividing the neural network into modular components. When a prompt is inserted into the form, only relevant elements are activated, skipping unnecessary paths. This selective activation reduces wasted processing power, and works in a similar way to the human brain.

The million-item input window opens new horizons for building advanced AI applications. By providing a wide scope for absorbing information, the professional appears to be able to “learn by doing,” that is, absorb new skills simply by incorporating evidence, without formal training. At the same time, the broad scope of information integration provides scope for dealing with complex challenges, which require significant practical knowledge.

Gemini 1.5 is an important step for Google in trying to balance innovation in developing AI models with their responsible management. Although it scores performance that rivals that of the high-end Ultra version, the Pro limits excessive demands on resources. As models evolve, minimizing unwanted side effects remains a priority. By focusing on target objects, Pro sets new limits on what AI can achieve while avoiding excessive resource consumption.

This Pro beta is a first look at how Ultra will evolve and beyond. Google is gradually adjusting form size and performance through a tiered approach. By making the model accessible to more and more users, it enables them to benefit from the rapidly evolving AI language model.

Through gradual releases of Gemini to users, Google aims to make the AI ​​model it is developing useful for a range of human activities. Improvements included in version 1.5 include new contextual windows for developers to create advanced applications. At the same time, selective activation of models allows them to focus on specific queries, without wasting resources. As language models evolve rapidly, Google aims to make them more powerful and useful, while keeping the risks that accompany such a technological leap under control.