Google on Wednesday released new artificial intelligence (AI) models that outside developers can make their own, following similar moves by Meta Platforms and other companies.
The Alphabet subsidiary said individuals and businesses can build artificial intelligence software for free based on a new family of “open models” called Gemma. The company said it is making key technical data public, such as so-called model weights.
The move could lure software engineers to build on Google’s technology and encourage use of its newly profitable cloud unit. The company says the models are “optimized” for Google Cloud, and cloud customers who use them for the first time will receive a $300 credit.
Google isn’t making Gemma fully “open source,” meaning the company may still be involved in setting the terms of use and ownership. Some experts say open source AI is ripe for abuse, while others support this approach to expand the group of people who can contribute to and benefit from the technology.
With this announcement, Google hasn’t opened up its larger, more advanced Gemini model as it did with Gemma. The size of Gemma’s model is said to be 2 billion or 7 billion parameters, or the number of different values that the algorithm considers when generating output.
Meta’s Llama 2 model has parameter sizes ranging from 7 to 70 billion. Google has yet to reveal the dimensions of its largest Gemini model. By comparison, the GPT-3 model announced by OpenAI in 2020 has 175 billion parameters.
Chipmaker Nvidia said on Wednesday it had worked with Google to ensure the Gemma model runs smoothly on its chips. Nvidia also said it will soon develop chatbot software to work with Gemma, which is being developed to run artificial intelligence models on Windows PCs.
(Except for the headline, this story has not been edited by NDTV staff and is published from a syndicated feed.)