Nilekani received accolades from Aravind Srinivas, writing in an X (formerly Twitter) post, for his contributions toward technological advancements in India–a tremendous accomplishment, to which Nilekani’s recommendations to Indian AI startups to keep their focus only on building practical AI applications rather than collaborating to develop large language models has come under attack.
Indian-origin Perplexity AI head Aravind Srinivas blasted Infosys co-founder Nandan Nilekani over his comments regarding AI, saying India should focus on both training model designs and practical AI deployment.
Tata Communications Aims for AI and Strategic Growth with the Launch of GPU-as-a-Service and Investment in NetFoundry
“As they define their use cases and investigate AI capabilities, we are aiming to reach a variety of startups and enterprises, including manufacturers and financial institutions,” Lakshminarayanan stated.
“To be clear: Nandan Nilekani is awesome and has accomplished more for India with Infosys, UPI, and other initiatives than any of us could have ever imagined. However, his suggestion that Indians should disregard model training abilities and instead concentrate on expanding upon pre-existing models is incorrect. Both are necessary. Srinivas wrote.
Srinivas’ remarks were a reaction to Nilekani’s previous counsel to Indian AI entrepreneurs. During the October Meta AI Summit, Nilekani advised companies to focus their resources on developing useful AI solutions rather than the expensive task of developing massive AI models.
“Our aim should not be to develop one more LLM. Let the big boys in the (Silicon) Valley do it, burning billions of dollars. We would use this as a way to create synthetic data, create small language models in a flash with an appropriate dataset and train them,” said Nilekani. He stressed the need for a scalable, affordable infrastructure and practical applications.
Srinivas, however, had a different opinion. He called for a more ambitious approach to AI development in India. Drawing from his own experience, he stated that India is committing a strategic error by not training AI models under the misguided assumption that the costs would be prohibitively large.
“I think India made the same mistake I did when I was running Perplexity. “Training thinking models is going to be very expensive,” Srinivas said. In order to be internationally competitive across all benchmarks, including Indic languages, he advised Indian entrepreneurs to stop depending on open-source models and instead concentrate on building their own AI skills.
Srinivas reiterates that India should draw parallels between its capabilities in AI with its achievements in the arena of space exploration, drawing an example from ISRO for overcoming resource constraints to gain world acclaim, with appreciation from SpaceX CEO Elon Musk, who finds missions of ISRO which have tasted success at a fraction of the cost and are efficient.
“Elon Musk appreciated ISRO (not even Blue Origin) because he respects those who can do stuff without overspending. That’s how he runs his business,” Srinivas wrote in encouragement to adopt the same mindset for AI in India.
He expressed hope at the close for a change in the direction of India’s AI vision. He commented on the development of a pool of AI experts to train big AI models capable of functioning globally, with proper local considerations. He extended his help to anyone who is “crazy enough” to take on this completely open-source-building challenge for wider benefit.