Global AI Conclave: LLMs vs applications? Lightspeed, Premji Invest, Accel and Peak XV Partners weigh in

Even Prayank Swaroop, Partner at Accel, agreed with Peak XV's Lakhani and Premji Invest' Pahwa and said LLMs are only doing what is done, there’s nothing new currently.

At a time even tech leaders, founders and investors are divided between India joining the large language models (LLMs) race or should it focus on building applications, the country’s top fund managers weighed in on the debate while speaking at the CNBC-TV18 & Moneycontrol Global AI

Conclave 2024.

Peak XV Partners’ managing director, Shailesh Lakhani, said, “There are so many new things possible with AI and applications so for most founders that seems like the most straightforward place to start. In applications, there are so many ideas from grating vegetables to processing an insurance claim faster. There’s a zillion problems to solve and we should have way more people trying to do those things today.”

Similarly, Manjot Pahwa, head of early stage technology investments at Premji Invest, said, “The most obvious thing to start with is applications. There are too many picks and shovels and barely any applicants to be using those picks and shovels.”

“The next generation of companies I am really excited about is the application layer. There are a lot of revenue pools and efficiencies that can be gained by the sensible application of AI. We should learn to leverage some of our strengths (services) and continue to build things that are easily winnable in the short term,” she added.

Even Prayank Swaroop, Partner at Accel, agreed with Lakhani and Pahwa and said LLMs are only doing what is done, there’s nothing new currently.

“People who are finding what is going to happen seven years from now and trying to invest in that is what I am going to find. Whether it’s going to be LLMs, applications or some other new technology, that’s what we’re searching for. Otherwise, we’re just playing second fiddle to Silicon Valley all the time,” he said.

Hemant Mohapatra, Partner, Lightspeed India said India has the data sets available to build the next big foundational model. From the audio data available with T-Series to the large video corpus that Bollywood owns, it should be relatively easy for India to build its own foundational models.

“India should be building these models but we’re not seeing enough of it happening. We don’t have the talent existing in the market today, we can pull from other markets but we don’t have that talent today,” Mohapatra said.

“My advice to founders is play the wars you are capable of winning today. If you’re amazing at raising billions of dollars today, go raise that money and build that model. But, if you’re not, that’s a big boy’s game,” he added.

Peak XV’s Lakhani also agreed that building LLMs is very capital intensive and may not be for everyone. “The global race for models may only be won by hyperscalers who have large free cashflow machines and other businesses that let them invest in (LLMs)...If scaling laws are truly holding and you need $100 billion…that may just not be available in the private markets for a startup company,” Lakhani concluded.