Google has expanded its Gemma 2 family by introducing new AI models focused on security and transparency. These new models, named Gemma 2 2B, ShieldGemma, and Gemma Scope, promise greater openness compared to other alternatives.
Gemma 2 2B, ShieldGemma, and Gemma Scope The modelsโGemma 2 2B, ShieldGemma, and Gemma Scopeโare built on the foundation of the Gemma 2 model, which Google launched in May. Gemma 2 2B stands out as a lightweight model capable of running on devices like laptops and tablets. ShieldGemma was developed to detect inappropriate content such as hate speech, harassment, or obscenity in both user inputs and the modelโs responses. Known as security classifiers, this model helps filter out unwanted responses.
The latest member of the family, Gemma Scope, offers developers the chance to explore how the Gemma 2 model processes information and makes predictions. Special neural networks embedded in Gemma Scope make it easier to analyze the modelโs commands.
Google has stated that these models are licensed for specific commercial and research uses and can be downloaded from resources like the Vertex AI platform and Kaggle for data science. By introducing these new models, Google is aligning with the U.S. governmentโs push for more transparent AI models. While ongoing monitoring for risks remains necessary, the Gemma 2 models are expected to address many of these concerns.
