Google unveils safer, smaller, more transparent AI models

Multinational technology company, Google has launched three new generative AI models, which it claims to be “safer,” “smaller,” and “more transparent” than most others.

The new Google AI models, part of the Gemma 2 family, include Gemma 2 2B, ShieldGemma, and Gemma Scope, each designed for different applications focusing on safety.

The Gemma series distinguishes itself from Google’s Gemini models by being open-source, aimed at fostering goodwill within the developer community, similar to Meta’s Llama.

Read Also: 

Google DeepMind showcases robot navigation with Gemini AI

Apple launches AI features in iOS 18.1 developer beta

Gemma 2 2B, a lightweight model, is designed for text generation and analysis, compatible with various hardware, and available for research and commercial use through platforms like Google’s Vertex AI, Kaggle, and AI Studio.

ShieldGemma focuses on safety, equipped with classifiers to detect toxicity, including hate speech and explicit content. It filters both prompts and generated content to ensure safety. Gemma Scope, on the other hand, allows developers to dissect the inner workings of Gemma 2 models, making their processes more interpretable.

The release aligns with the U.S. Commerce Department’s recent endorsement of open AI models, emphasizing their importance for accessibility to smaller entities while acknowledging the need for monitoring to mitigate risks.