Avoid costly, large LLM buildouts; focus on smaller models: Sridhar Vembu on India AI strategy – Crypto News

New Delhi, Feb 8 (PTI) India should not emulate or compete head-on with the massive Large Language Models (LLMs) currently dominating the AI landscape, Zoho founder and Chief Scientist Sridhar Vembu has said, advocating that the country should focus on smaller models or other approaches that are less-energy, less capital intensive.

The remarks come at a time when interest in artificial intelligence (AI) has peaked in India, with the country soon to host the India AI Impact Summit – the largest of the four major global AI congregations so far.

Vembu’s view also aligns with the recent Economic Survey, which noted that India’s limited access to cutting-edge compute infrastructure and capital makes pursuit of foundational models as the centrepiece of an AI strategy rather “challenging”.

The Survey favoured a bottom-up approach to AI development saying it aligns more closely with realities.

In an interview with PTI, Vembu cited the USD 50-100 billion costs and high-energy demands of large models built by the ‘Big Tech’ – a term used for the world’s largest and most influential technology companies that dominate digital platforms, data, and innovation.

Vembu noted that in this context, “Sometimes, staying a little bit behind is a good idea.”

Rather than emulating energy-hungry global models, he argued that India should leverage its “brain power” to invest in research and development around smaller, more efficient AI approaches.

“If you look at AI today, the big tech…the big LLMs dominate. That is three or four well-known names… the Chinese are in the game from an open source model perspective…But if you look at the domination, it is from the big LLMs,” he elaborated.

He believes India should avoid entering the capital-intensive LLM race and instead pivot to smaller, distinct approaches.

“I don’t think we should take that (large LLM model) head-on right now, because this is a USD 50 billion, 100 billion game…the GPUs are in short supply, and are expensive. And the energy intensity of this, the electricity prices are going up rapidly in the US. So I don’t believe we should simply emulate that right now. Sometimes, staying a little bit behind is a good idea,” he pointed out.

Vembu said there is significant scope for research and development in AI, and added that Zoho is actively pursuing more energy-efficient routes, including smaller models or other approaches.

“There are other approaches to AI that don’t involve this type of large language model approach. Those all are promising, and we are pursuing some other directions. We have to invest in that, we have to apply our brain power, rather than energy which is scarce,” said Vembu, a prominent voice on technology, innovation and the societal implications of artificial intelligence.

The debate comes as India gears up to host the India AI Impact Summit later this month. Billed as the largest of the four global AI summits hosted so far – following the AI Safety Summit hosted by the UK, the AI Seoul Summit, and the AI Action Summit hosted by France, the event reflects growing international focus on responsible, inclusive and impact-driven AI, as well as India’s expanding role in shaping the global AI narrative.

According to an official release earlier last week, the Summit has garnered strong interest from the global community with over 35,000 registrations received ahead of the event, at the last count.

Governments, industry leaders, researchers, civil society organisations and international institutions are set to actively participate in shaping the Summit’s agenda. It is expected to see participation from over 100 countries, including 15 to 20 Heads of Government, 50 plus Ministers from various countries, and 40 plus prominent global and Indian companies.

Around 500 leading names from the global AI ecosystem, including innovators, researchers and Chief Technology Officers, are expected to attend the mega event which will feature a showcase of over 500 AI startups and host around 500 sessions alongside the main programme.

The event will be closely watched globally, with several influential figures shaping the global tech discourse slated to participate. That list includes Jensen Huang, Founder and CEO, NVIDIA; Dario Amodei, Chief Executive Officer (CEO), Anthropic; Brad Smith, President and Vice Chair, Microsoft; Demis Hassabis Co-founder and Chief Executive Officer (CEO), Google DeepMind; and Julie Sweet Chair & CEO, Accenture among others.