Google launched Gemma 2 a few days ago, the new version of its family of "open" artificial intelligence models. This update does not replace the Gemini models but will coexist with them, offering new capabilities and greater flexibility. The first version of Gemma was released in February, and now, in August, comes this second iteration.

 

What is Gemma 2 and what is it for?

Gemma 2 is a family of artificial intelligence models designed to be highly adaptable. Instead of being a single model, Gemma 2 is a platform with various variants that have specific characteristics, allowing their use in different contexts according to the user's needs.

While Gemini focuses on large-scale and "open" models, Gemma 2 is designed to offer pre-trained models that developers can modify according to their needs. Although Google describes these models as "open," there are certain restrictions in the license terms. However, they allow enough versatility for developers to adapt them to a wide range of applications.

The flexibility of Gemma 2 means that developers can select and adjust the specific features they need. Originally created for natural language processing in English, its design allows for additional training to suit specific needs. In addition to generating text from prompts, Google's tools allow training Gemma 2 for other tasks, such as code generation and solving mathematical problems.

 

New Models in the Gemma 2 Family

The new version of Gemma 2 includes three main models that Google offers to its clients:

  • Gemma 2B: This is the lightest and most versatile model, designed to generate text flexibly. It can be used on a wide range of devices, including personal computers and peripherals.
  • ShieldGemma: A security model that acts as a defense against harmful content. It can detect harassment, hate speech, and explicit sexual material, allowing the filtering of user requests to generate content via AI.
  • Gemma Score: This model evaluates the performance of other models in the Gemma 2 family. Its specialized neural networks review the information processed by Gemma 2, facilitating the continuous research and improvement of Google's artificial intelligence.

Gemma 2 represents a significant advance in Google's family of AI models, offering a solid and adaptable base for developers and researchers. With these new tools, progress in the field of artificial intelligence continues, providing innovative solutions for a wide variety of technological challenges.