Google Has Updated Gemma With Safe AI Models

The other day, Google announced the launch of three new models of artificial intelligence: Gemma 2 2B, ShieldGemma, and Gemma Scope. These models are distinguished from others by increased security, compactness, and transparency. They are part of the Gemma 2 range, which was introduced earlier this June. What sets the Gemma series apart from Gemini is its accessibility to developers. This will allow Google to strengthen its relationship with the development community. By the way, Meta has a similar goal, which it implements in its Llama model.

Gemma 2 2B is a simplified model for writing texts that can work on various devices, including laptops and mobile phones. It is licensed for research and commercial use. The novelty is compatible with PyTorch and TensorFlow libraries. It can be accessed through platforms such as Vertex AI Model Library, Kaggle, and Google's AI Studio. You can also use the new model through the Hugging Face platform. Gemma 2 2B has only 2.6 billion parameters. Despite its relatively compact size, it is able to compete with much larger models thanks to its advanced learning technology and high-quality data.

ShieldGemma consists of a set of so-called "safety classifiers" that detect unwanted content, including hate speech, stalking, and sexual images. This tool is based on Gemma 2 and can be used to filter incoming prompts and content generated by the model itself.

Gemma Scope allows developers to better understand the internal processes of Gemma 2. This model uses specialized networks to simplify complex data, making it easier to analyze.

Google says the new open AI models of the Gemma 2 series will make it easier for small firms, researchers, non-profits, and individual developers to use generative AI. At the same time, the company emphasized the importance of monitoring these models in order to identify possible risks and avoid them.