This move towards a more open AI model is unusual for Google, which typically leans towards closed access, making it a contentious figure in the AI debate over open and closed research. However, Google stopped short of fully open sourcing Gemma, instead referring to it as an “open model.” This means the model's pre-trained parameters are available, but not the actual source code or training data. The Gemma model will be available for "responsible commercial usage and distribution for all organizations, regardless of size."
Key takeaways:
- Google DeepMind has unveiled its latest set of AI models, Gemma, which are open for commercial use and redistribution by AI developers and researchers.
- Gemma, which comes in two sizes, Gemma 2B and 7B, can run directly on laptops and desktop computers and is based on the research and tech used to build Gemini, Google’s marquee language model.
- Despite being a departure from Google's usual closed AI technology, Gemma is not fully open sourced. The model's pre-trained parameters are available, but not the actual source code or training data.
- Google's move towards a more open model with Gemma may suggest a broadening strategy for the company, which has been criticized for its closed approach to AI technology by open source proponents.