Google Unveils Open Gemma AI Models as Meta Competitor

Google has introduced a new set of artificial intelligence models known as Gemma, aimed at developers and researchers in the field. These models are part of a broader trend in the tech industry toward more openness in AI technology. While Google is making technical data such as model weights available to the public, it has not released the source code or training data. The Gemma models are smaller in comparison to some competitors, with sizes ranging from two billion to seven billion parameters, and are designed to be more accessible, with the capability to run on laptops as well as high-capacity AI chips and data centers.

The release of Gemma is seen as a strategic move by Google to boost the usage of its cloud division, Google Cloud, and to foster a community of software engineers around its AI technology. Despite this step towards openness, Google's premier models, like Gemini, remain closed to the public. This selective openness indicates that Google may retain a degree of control over the terms of use and ownership of Gemma.

Alongside Gemma, Google is offering tools to ensure responsible usage, including a Responsible Generative AI Toolkit and a debugging tool to explore the model's outputs. The models are available for free on the Kaggle platform, encouraging transparency and community validation of their safety. Meanwhile, Nvidia has announced plans to integrate Gemma with its chips and develop chatbot software compatible with Windows PCs, suggesting a growing ecosystem around these open AI models.