Editorial: Homegrown AI platform
The government funding support is said to be over Rs 200 crore, which will mostly be in access to the scarce and expensive dedicated computing infrastructure.
Representative Image
After all the hullabaloo over Commerce and Industry Minister Piyush Goyal’s criticism regarding Indian startups, the government seems to have received the right prompt from the startup ecosystem and has responded with a correct solution. The Centre will be offering generous support to the Bengaluru-based startup Sarvam AI to build “India’s first home-grown sovereign Large Language Model (LLM)”. LLM is a type of artificial intelligence model, and sovereign LLM generally refers to a nation’s capability to develop and deploy AI using its infrastructure and datasets. For India, it is a small step forward even as the US and China have been making giant leaps.
Helping build its own AI model has been on the government’s agenda for a while now. Last March, the government approved the setting up of the India AI Mission, and in January this year, it called for proposals to collaborate in building a sovereign LLM.
The government funding support is said to be over Rs 200 crore, which will mostly be in access to the scarce and expensive dedicated computing infrastructure. Sarvam has, on its own, raised about $41 million from venture capitalists in December 2023. So, the real value of government support, besides giving it a shot in the arm and raising its credibility, will be in getting clearances and access to national datasets that will expedite and fast-track development. There are still concerns regarding the data governance framework and laws, which are still in a nascent stage. The National Data Governance Framework Policy and the Digital Personal Data Protection Act (DPDP) of 2023 are perceived to be works in progress as the domain has been undergoing continuous changes, and criticism from well-meaning quarters indicates that the DPDP Act has room for improvement.
The unique thing about the Indian LLM will be customisation to the requirements of a diverse country which has multiple languages and varying levels of literacy and computer literacy, despite a change in the latter due to widespread access to smartphones across all sections of the society.
Sarvam models have been designed to support Indian languages and, more importantly, enable voice interactions.
The path to building the indigenous LLM is full of thorny challenges. The two most obvious ones are capital and time. LLM projects are money and resource guzzlers and take an inordinately long time.
Moreover, setting up the necessary computing infrastructure – that is, supercomputers and HPC or high-performance computing – requires huge investments, and there’s a risk of geopolitical supply chain challenges arising out of the ongoing global trade conflicts. Thus, building a robust model and its countrywide and population-scale deployment would need much longer timelines. Another major constraint is the availability of a skilled workforce. The AI talent pipeline needs to be put in place, and that would require sweeping changes in engineering and science education to incorporate cutting-edge technologies relating to AI, machine learning and data in the curriculum in a structured way.
There is an “India Future Skills” initiative, but it has a long way to go.
Despite being an IT and software services powerhouse for nearly three decades, India has not had much success in developing global technology products and platforms that could match and compete with those of American and Chinese tech giants. The start-ups could draw inspiration from the development and deployment of UPI in India as a public-private partnership, which accounts for half of the global real-time payment transactions.
