Google unveils new generative AI models and benchmark tool tailored for Indian languages
At its recently concluded I/O Connect event in Bengaluru, tech giant Google announced several updates to its existing large language models (LLMs) along with the release of a new benchmarking tool specifically designed for Indic languages.
Addressing the event, Ambarish Kenghe – VP of Google India, emphasized the company’s commitment to empowering Indian innovators through harnessing the full potential of AI. He stated that opportunities in multi-modal, mobile and multi-lingual AI are immense and that Google is pleased to support India’s AI journey.
Google revealed the latest versions of its popular generative models Gemini and Gemma. Developers can now access an expanded context window of up to 2 million parameters on Gemini 1.5 Pro. A new model called Gemma 2 with a gigantic 27 billion parameters was also unveiled.
In a first, Google launched IndicGenBench – a benchmark tool aimed at evaluating how well an AI model performs on Indian languages. This will help developers test and improve their language models.
It was highlighted that over 1.5 million developers globally are already leveraging Gemini models integrated across Google products like Chrome, Gmail and Maps. The Nano version optimized for mobile will debut on select Pixel phones later this year.
Meanwhile, Google DeepMind’s Project Vaani was expanded in collaboration with IISc Bangalore. It now provides over 14,000 hours of multi-lingual speech data collected from 80,000 speakers across India for use in AI development.
Emphasizing access for Indian talent, subsidized pricing for Google Cloud APIs were announced along with $350,000 credits for local innovators. Various initiatives to partner with organizations like ONDC were also detailed.
