Meet Gemma – Google’s Latest Innovation in Easy-to-Use AI Technology for Developers

Google is now offering Gemma 2B and 7B , a pair of open-source alternatives to the Gemini AI . These models provide more flexibility than Gemini and can integrate with local or cloud-based services. However, the Gemma platform is less powerful than Gemini, so it may be limited to simple tasks.

As you may know, Google Gemini is freely available on the web and requires no technical expertise. These new Gemma models aren’t intended for casual use—they’re specifically made for developers and researchers who want to work with open-source conversational AI. If a developer wants to build an app-in chatbot, for example, they may do so using Gemma 2B or 7B.

Developers can integrate the more powerful Gemini model with apps or services. But this integration requires cloud-based, closed-source platforms like the Gemini API and Google Vertex AI. Gemma, on the other hand, can run locally or within the cloud. Plus, its open-source nature is ideal for some specialized applications, as developers have more freedom to modify and customize the AI’s behavior.

Open-source AI also promotes transparency and may be used as a learning or research tool. That said, there are some concerns associated with open-source AI models like Gemma. The most obvious concern, at least in 2024, is that malicious actors may use open-source AI to drive misinformation campaigns on social media.

Google says that the Gemma models underwent extensive “red-teaming, automated adversarial testing, and assessments of model capabilities for dangerous activities.” In some cases, these safety tests were more rigorous than the tests endured by Gemini. The AI models ship with “responsible AI toolkits” that allow developers and researchers to create additional guardrails. A debugging tool is also included. Additional safety details are listed in the Gemma Model Card .

To reiterate, the open-source Gemma models are less powerful than Gemini. But it may be more powerful than other open-source models, such as Meta’s Llama 2 7B , which failed to beat Gemma 7B in benchmarks conducted by Google.

Developers and researchers can begin using Gemma today. The model is authorized for commercial use and distribution, though there are some clauses in its terms of use . Note that the Gemma models are currently limited to English-language applications.

Source: Google