In an effort to improve the safety and transparency of AI, Google recently launched three new generative AI models. The models are part of Google's Gemma 2 series, which are designed to be safer, more efficient, and more transparent than many existing models.
The company wrote that the new models - Gemma 2 2B, ShieldGemma, and Gemma Scope - build on the initial Gemma 2 series launched in May.
Unlike Google's Gemini model, the Gemma series is open source. This approach reflects Meta's strategy, whose Llama model aims to provide accessible, powerful AI tools to a wider audience.
Gemma 2 2B is a lightweight model for generating and analyzing text. It is general enough to run on a variety of hardware, including laptops and edge devices; meanwhile, the ShieldGemma model focuses on improving safety by serving as an ensemble of safety classifiers for detecting and filtering toxic content, including hate speech, harassment, and explicit sexual content. It runs on top of Gemma 2, providing a content moderation layer that is critical in today's digital environment.
The Gemma Scope model allows developers to gain deeper insight into the inner workings of the Gemma 2 model, and consists of specialized neural networks that can help unravel the dense, complex information processed by Gemma 2. (Cointelegraph)