Infosys chairman Nandan Nilekani: ‘Let the big boys in the Valley build LLMs; we will use it to solve real-world problems’

Infosys chairman Nandan Nilekani (R) with Meta's chief AI scientist Yann LeCun (Centre) and People+AI's Tanuj Bhojwani (L)

Infosys chairman and Aadhaar architect Nandan Nilekani remains steadfast in his vision of establishing India as the artificial intelligence (AI) use-case capital of the world, rather than focusing on the hypercompetitive race of building Large Language Models (LLMs).

"Our goal should not be to build one more LLM. Let the big boys in the (Silicon) Valley do it, spending billions of dollars. We will use it to create synthetic data, build small language models quickly, and train them using appropriate data," Nilekani said at Meta's Build with AI summit in Bengaluru on October 23.

Nilekani said their focus will be on creating infrastructure for collecting appropriate data. "It's all about data. How do we create the infrastructure for collecting the right data and make India the use case capital of AI globally where we actually deploy, add scale and speed in a frugal manner," he said.

"Let other people build LLMs, we will make sure it works for people" he said.

He also projected that we will have "high quality, accurate inference at very low cost and population scale in the next 12 to 18 months".

These remarks come a few months after Nilekani outlined a vision of putting AI in the hands of people in India.

"The Indian path in AI is different. We are not in the arms race to build the next LLM, let people with capital, let people who want to pedal chips do all that stuff... We are here to make a difference" he said at an event hosted by People+AI in Bengaluru in May 2024.

‘Meta's Llama open-source move a game changer’

During the summit on October 23, Nilekani also commended Facebook parent Meta's efforts to open source Llama, the social networking giant's collection of foundational large language model (LLM), terming it as a "game changer for us in India and something we need to take full advantage of".

In July, Meta changed the licensing terms for Llama AI models alongside the launch of the Llama 3.1 model. Developers can now use synthetic data generated by Llama models to create derivative models or train other models. This change is expected to give developers a lot of flexibility in creating custom smaller models that are optimised for their use case.

In September, Meta released the Llama 3.2 model with multimodal capabilities, which means it can understand different formats such as text and images at the same time. It is available in four variants - small and medium-sized vision LLMs (11B and 90B parameters), and lightweight, text-only models (1B and 3B parameters).

Read: 'Mark Zuckerberg will go down in history for...': RIL chief Mukesh Ambani hails Meta founder

"Open source has been key to the digital public infrastructure work in India...I see no choice but to use open source," he said.

Nilekani believes that today's AI technology can solve a large number of problems in areas such as agriculture and education at a population scale. It can also augment India's Digital Public Infrastructure (DPI) and accelerate inclusion across the country, he said.

"We feel that we have a unique position in India where, for the past 15 years, we have worked on building DPI at scale. We have learnt what it means to build technology at that scale, how to design it to be inclusive, innovative and allow market innovation to happen. Therefore, we think that we can build on top of DPI. DPI actually is giving us a foundation to go to AI much more rapidly" Nilekani said.

"The idea is how do we remove bottlenecks. For example, if we can create a digital public good of Indian languages, everyone can use it, whether it is a startup, government, or large companies," he added.

Balance between responsible AI and innovation

Nilekani also emphasised the need to balance responsible AI with fostering innovation.

"I have spoken to some of our leaders, and I think they clearly understand that we don’t want to have a repressive approach to AI innovation; we don’t want to create such heavy rules that you can't innovate," he said. "I think we should go full speed ahead while ensuring guardrails are in place."

Meta's VP and Chief AI Scientist, Yann LeCun, who is often referred to as one of the Godfathers of AI, also backed Nilekani's view.

In an interview to Moneycontrol, he said that using an open-source foundation model would be the right approach for India at present, which can be fine-tuned to deploy for a variety of vertical-specific applications across both consumer and business domains.

"We need to have foundational models that speak all the languages in the world and a single entity like Meta does not have access to all the data in the world, in part because regions want some sovereignty over their own data and control," he said.