Google has announced a new family of “open” AI models. The company says Gemma 2 models are designed to promote responsible AI development and are available for use by developers and researchers worldwide. The release includes two model sizes, Gemma 2B and Gemma 7B, each available in pre-trained and instruction-tuned variants.
Gemma models are not fully open source, but Google is making key technical details (such as model weights) publicly accessible. This lets developers build their own AI applications while still providing some control over the models’ usage and distribution. In addition to the models, Google also announced its Responsible Generative AI Toolkit, which includes guidance and tools for creating safer AI applications.
We’re still testing Gemma 2 2B and 7B, so I can’t comment on Google’s claims that Gemma 2 2B benchmarks about as well as GPT 3.5, but the specifics are less important than the trend. We are going to see very capable small models that will run well on smartphones, tablets, and laptops. Imagine how consumer behaviors will change in a world where AI can run locally (no network or cloud required) on any device restricted only by required outcomes (business rules). That’s what’s coming. The trend is clear.
Author’s note: This is not a sponsored post. I am the author of this article and it expresses my own opinions. I am not, nor is my company, receiving compensation for it. This work was created with the assistance of various generative AI models.