Share to FacebookShare to TwittShare to Linkedin
On Wednesday, Google Deepmind introduced its latest series of AI models named Gemma, marking a shift towards openness in the realm of AI technology. Unlike their previous closed-off approach, these models are now accessible for commercial utilization and sharing among AI developers and researchers, aligning with the growing trend of collaborative AI research.
Gemma, a collection of lightweight models, draws inspiration from Gemini, Google’s prominent language model that competes with OpenAI’s ChatGPT. The distinguishing feature of Gemma lies in its versatility, offering two variations: Gemma 2B and Gemma 7B, tailored to run efficiently on laptops and desktops. This accessibility enables developers and researchers to leverage these models akin to Meta’s LLaMA 2 model.
Google’s decision to embrace a more open model like Gemma contrasts with its historical preference for closed systems, sparking discussions within the AI community regarding the merits of open versus closed research approaches. While proponents of open collaboration, such as Meta and Hugging Face, advocate for transparent sharing of source code to foster scientific advancement, Google and OpenAI lean towards a controlled environment to address safety concerns and prevent misuse.
Although not fully open-sourced, Gemma is positioned as an “open model,” offering access to the model’s pre-trained parameters while keeping the source code and training data proprietary. This distinction sets a precedent for Google, which has faced criticism for its restrictive practices, prompting speculation about a potential shift in strategy.
In contrast to the limited accessibility of Gemini 1.5, which was restricted to developers and Google’s cloud clients, Gemma’s inclusive approach signifies a departure towards greater openness. This strategic shift may signal Google’s response to criticisms from proponents of open-source initiatives, notably Meta, who have raised concerns about impediments to progress in the AI field.
Demis Hassabis, the leader of Google DeepMind, has defended the company’s closed approach, emphasizing the need to address risks associated with AI systems before widespread deployment. The launch of Gemma also includes an API with resources and guidelines to facilitate safe utilization of the model, underscoring Google’s commitment to responsible AI development.