Global AI Conclave: Capital needed for Robotics and AI startups huge, India doesn’t have that kind of funds today, says ARTPARK co-founder Umakant Soni

Need for building foundational models

Both Soni and Bharadwaj Amrutur, the executive director of ARTPARK and professor at IISc spoke about the need for building foundational large language models (LLMs) in India, even as existing LLMs from global tech giants continue to provide a huge opportunity to build applications over it and is still untapped.

This comes at a time when some of the biggest tech world voices in India including Infosys co-founders Narayana Murthy and Nandan Nilekani have said that India should focus on building use cases instead of trying to develop another LLM after spending billions of dollars.

Amrutur explained, “LLM is actually a repository of knowledge which then provides it in an actionable form. If you don’t have control over it and if you start using that for your applications to take decisions, you are relying on someone else for curating that knowledge. This could be okay for a bulk of things but there could be a lot of specific strategic use cases where you don’t want to lose that control. Then there is of course the issue of bias.”

He also noted another interesting area of bringing physical intelligence in Robotics that is yet to happen. “Creating foundational models in this area, LQMs (large quantitative models), is a huge opportunity where we should not be left behind. A lot of this data will come from simulations and computations like scientific AI, calculations, giving data etc.”

Soni too addressed the risks of not having country’s own LLMs or foundational layers. Citing examples of “level 3” and “level 4” cars moving 1.4 billion Indians, he said there could always be the risk of the vehicles going rogue or facing technical glitches. And not having access to foundational models could mean risking data leaks and control over such incidents.