Google releases open models of AI after Meta

Following similar moves by Meta Platform and others, Google on Wednesday released new artificial intelligence (AI) models that outside developers can potentially build into their own.

The Alphabet subsidiary said individuals and businesses can create AI software for free based on a new family of “open models” called Gemma. It said the company was making key technical data, such as what are called model weights, publicly available.

The move could attract software engineers to build on Google’s technology and encourage use of its new profitable cloud division. The company said the models are “optimized” for Google Cloud, where first-time customers using the cloud get a $300 credit.

Google stopped short of making Gemma completely “open source”, meaning the company can still have a hand in determining the terms of use and ownership. Some experts have said that open-source AI is ripe for abuse, while others have supported the approach of broadening the group of people who can contribute to and benefit from the technology.

With the announcement, Google did not open source its larger, flagship models known as Gemini, unlike Gemma. It states that the size of a Gemma model ranges from two billion or seven billion parameters – or the number of different values ​​that an algorithm takes into account to generate an output.

Meta’s Llama 2 models range in size from seven to 70 billion parameters. Google hasn’t revealed the size of its largest Gemini model. For comparison, OpenAI’s GPT-3 model, announced in 2020, had 175 billion parameters.

Chip maker Nvidia said on Wednesday that it has worked with Google to ensure that the Gemma model runs smoothly on its chips. Nvidia also said it will soon make chatbot software it is developing to run AI models on Windows PCs will work with Gemma.

(Except for the headline, this story has not been edited by NDTV staff and is published from a syndicated feed.)