Many clients seeking tailored Small Language Models, says Infosys AI head

Infosys

Information technology giant Infosys, on November 19, revealed that many of its customers have approached the company to develop Small Language Models (SLM) tailored to their needs.

“Infosys just launched two small language models about a month back. And a lot of our customers are saying, can you develop it for my context," said Balakrishna D.R, head of AI at Infosys, at the 27th edition of the Bengaluru Tech Summit 2024.

On October 24, Infosys announced the launch of two SLMs built using the NVIDIA AI stack, designed for banking and IT-specific applications. These models leverage industry-specific data in collaboration with Sarvam AI.

SLMs are machine learning algorithms trained on a more specific dataset than large language models (LLMs). The IT major said it plans to make foundational SLMs available as a service to businesses that can build solutions on top of them.

The debate between SLMs and LLMs in India has gathered momentum in recent times, centred around cost and accessibility for local needs. Additionally, SLMs are known for affordability while LLMs face criticism for high costs and privacy concerns.

“These are the opportunities for us that we can actually help the entire world out of AI, coming from home (India),” Balakrishna said, who is also the executive vice president, global services head, AI & industry verticals of Infosys.

From a services perspective, Balakrishna said that AI presents a significant opportunity for the Indian IT industry, given its enterprise customer base.

This is because the challenges enterprises face in adopting AI are multifold, including identifying the best use cases to maximise return on investment (RoI), managing internal data responsibly, and ensuring solutions remain sustainable amid rapidly changing technology.

“It is not the data that is there on the internet; it is data that they have in the enterprise. How would they actually leverage that? They have to figure out how to use this in a responsible manner,” he said.

Using Infosys as an example, Balakrishna highlighted that the company has developed industry blueprints for 23 different sectors, guiding enterprises on data strategies, responsible AI adoption, and scalable platforms. He likened this to the role Indian IT firms played in driving digital transformation across the globe.

“I think the Indian IT industries, many of them based in Bangalore, can actually help the entire world adopt AI. And I think that's where we see the opportunity and that's what we are actually working for.”

In August, Infosys chairman and Aadhaar architect Nandan Nilekani said India has put the world’s apprehensions to rest by building an SLM with two billion parameters.

Also read: Nandan Nilekani bats for data-driven small language models; bullish on Sarvam AI

Earlier, Nilekani, popularly referred to as 'India's CTO' in tech quarters, said that the country should focus on efficient and low-cost use cases of AI rather than obsessing about building bigger language models.