Ola cofounder’s AI venture keeps most details away for the future – Crypto News

The company will design its own chips over the next six months, which will be focused on efficiency, he claimed. However, Krutrim is yet to release any technical documentation on the model and details on the company’s approach towards industry adoption also remain unanswered as of now.

In a demonstration of the model, Aggarwal said that the eponymous AI model was trained on 2 trillion tokens—a size more or less in line with large foundational models released by global firms such as OpenAI, Meta and Google. He also added that a chatbot based on the Krutrim model can “write and speak in 10 Indian languages, and can understand 22.” Work on the model, Aggarwal added at a media roundtable post-announcement, had begun “roughly a year ago.”

“We’re also building our own technology for data centres—not just the physical investment into data centres, but also the technology. This is because it is important to make data centres more efficient to bring down cost, and to make it a greener, more sustainable solution,” he said. “In India, we need to design our own silicon chips for building this.” 

However, key details such as potential enterprise client conversations, the source of the data on which the models are being trained, and details on data centre and chip investments were not shared.

Speaking about the ownership of the venture, Aggarwal said, “Krutrim is a separate business altogether. It is not going to be integrated at a transactional level. There are some entities that I own 100%—this is under my company, and not part of Ola or Ola Electric’s corporate structure.” 

He also added that while Krutrim has “some investments into it,” the details were not publicly disclosed.

Aggarwal’s announcement comes amid a time of heightened interest in India-specific AI platforms as well as use cases. The centre is set to unveil its AI policy under the India AI programme on 10 January, which will include a policy framework for public-private partnership models on development of AI databases in Indic languages, as well as indigenous compute capacities, said union minister of state for IT, Rajeev Chandrasekhar, in an interview with Mint on Thursday.

The government, under the ministry of electronics and IT (Meity), already has Bhashini—a Centre-backed project under the Digital India initiative to create local language datasets and AI applications. Industry, researchers and academia have for long clamoured for access to better, more structured data and access to affordable compute on the cloud. On 16 April, Mint reported that these were crucial challenges that a number of top Indian universities continue to face in applied AI research projects.

Ola’s efforts to tap the AI market are likely an effort to address the government’s focus on AI, and growing industry needs.

Speaking about how Krutrim will tie in with Bhashini, Aggarwal said, “A lot of these projects are academia or research-oriented projects, rather than large scale commercial deployments. We look at a need for India’s own language model as very foundational, and to really build strong, capable foundational models needs a lot of investments into capital, compute bandwidth, and more, as well as a lot of core technology architectures, and a lot of data that we need to collect, process, ingest, and then create well-executed products for consumers. So, this is an effort that needs organizations, versus research efforts.”

Adding further on where the data, which every stakeholder has flagged as a challenge so far, was sourced from, Aggarwal said, “I can’t share my data sources publicly, but we have a lot of data operations efforts. We have proprietary datasets that we gathered over the last year. We have data partnerships, as well as open platforms that we sourced from. But it’s not just about the scale of data—it’s also about how you process the data.”

Industry stakeholders flagged what could be concerns in Krutrim’s initial unveiling. Kashyap Kompella, chief executive of tech consultancy firm RPA2AI Research, said that each area addressed by Aggarwal’s Krutrim “are capital-intensive, with long gestation periods.” He also added that if the venture seeks serious industry credentials, key steps would be to open-source the model, as well as release a technical whitepaper in line with industry standards.

A second senior industry consultant, who requested anonymity, further added that clarifications would be sought from Krutrim on its sources of data, and how it would intersect with Bhashini—which is already building AI use cases and applications. “Developing silicon design and data centres could cumulatively be billion-dollar efforts—opening them up right at the onset could be amply challenging,” the consultant added.

Milestone Alert!
Livemint tops charts as the fastest growing news website in the world 🌏 Click here to know more.

Unlock a world of Benefits! From insightful newsletters to real-time stock tracking, breaking news and a personalized newsfeed – it’s all here, just a click away! Login Now!

Published: 15 Dec 2023, 08:30 PM IST