At an event in Delhi in June this year, the CEO of OpenAI, Sam Altman, was asked if a young team of Indian engineers, loaded with $10 Mn, could build an LLM model similar to that of OpenAI.
Responding to the question, Altman said that it would be “hopeless to compete with ‘us’ (OpenAI) on training foundation models [and] you shouldn’t try”.
Altman’s comments triggered an immediate backlash and stirred discontent among the Indian tech community.
While the comment of OpenAI’s reinstated CEO is now history, it has started a catalytic reaction of sorts, as a wave of Indian founders is now seen floating projects dedicated to building foundational LLMs for India and the world.
It is imperative to mention that right after Altman’s infamous remarks, Sridhar Vembu, the founder of Zoho, announced plans to create a proprietary LLM with capabilities encompassing summarization, paraphrasing, and adaptability to novel tasks. Meanwhile, Tech Mahindra is working on Project Indus, an open-source Indic language-based LLM. Similarly, numerous AI startups, including Sarvam AI, SoketLabs, CoRover.ai, are actively engaged in developing LLM models with a focus on the Indic languages.
Adding to this wave of AI-centric innovation, Bhavish Aggarwal, the founder of Ola, has launched an AI company, Krutrim, which claims to have its focus set on constructing foundational LLMs for such languages.
As of now, Krutrim is proficient in understanding and processing up to 22 Indic languages with the ability to write and speak in 10 languages, as per Aggarwal.
At a time when the LLM fight seems to be intensifying worldwide and in India, Aggarwal-led Krutrim wants to be a step ahead in the AI game. It aspires to develop the entire cloud infrastructure and silicon chips, excluding the fabrication aspect.
Before we delve into Krutrim’s DNA, let’s understand the startup’s ownership and funding status.
Krutrim SI Designs operates as an independent entity and pursues funding autonomously. It has secured an undisclosed amount of funding from investors, including Aggarwal.
Although Aggarwal did not disclose its existing shareholding status, he confirmed that the company has already raised some funding. As per reports, the newly incorporated AI startup had raised $24 Mn in debt from Matrix Partners in October this year.
Meanwhile, sources close to the company informed Inc42 that Krutrim was in the process of securing another round of funds, with the announcement expected before March 2024.
Now, let’s steal a glance at the core of Krutrim’s DNA and understand how it could be the answer to the GPTs of the world in India.
Krutrim Wants To Build An India-Centric AI
Serial entrepreneur Aggarwal, who founded Ola Cabs in 2010 and Ola Electric in 2017, asserted that many popular international LLMs lack true representation of the Indian context and data.
“If they genuinely want to adhere to the democratic values, they must ensure that the data is representative of the population scale. About 20% of the data in these models should be Indian, reflecting democratic values. Unfortunately, this is not the case, leading to outputs that present a caricature of India, even in something as basic as an Indian woman’s saree,” Aggarwal said.
At Krutrim, the intention is to build multimodal models, particularly image generation models. “While we are not there yet, we initially focussed on text and language models. We are actively seeking partnerships within the ecosystem to bring our models to the market and collaborate in creating Indian data sources for these other modalities,” Aggarwal added.
Additionally, Aggarwal emphasised that most Western AI solutions not only lack accuracy in the Indian context but are also prohibitively expensive. Krutrim aims to address these challenges by developing full-stack AI solutions.
The startup is set to accomplish several ambitious goals, encompassing foundational LLMs, AI cloud infrastructure, and the in-house development and manufacturing of AI-optimised silicon chips.
The head of strategy at Ola Electric, Ravi Jain, emphasised that the company is actively designing and developing in-house technologies such as a submersion cooling system for cloud infrastructure and chip designs.
“Krutrim’s innovations will simplify and streamline everyday interactions across various socio-economic strata in India. For instance, tasks like reserving a table could become more straightforward. In the future, Krutrim aims to manage extensive customer support for emerging startups and businesses, providing real-time information and aiding in product selection, thereby impacting even rural economies significantly,” Jain said.
The base LLM of Krutrim is trained on an impressive 2 Tn tokens, representing the largest dataset of Indian data used in training to date, according to Jain.
He added that Krutrim is designed to strike the right balance between performance and price, making it suitable for powering a variety of day-to-day applications. Additionally, the company is developing Krutrim Pro, a large multimodal model with advanced problem-solving and task execution capabilities, slated for launch in the next quarter.
Krutrim is set to be available in beta version as an API for enterprises and developers starting January 2024, facilitating the creation of AI-driven assistants capable of conversing in multiple Indian languages. Krutrim Pro is anticipated to be available in Q4 FY 2024.
Aggarwal envisions Krutrim’s superior linguistic skills as a valuable tool across diverse domains, ranging from education to business communications. The platform incorporates the latest techniques in safe AI to mitigate inappropriate responses.
It is also actively working on developing AI infrastructure, including indigenous data centres, server computing, edge computing, and supercomputers, with production scheduled for mid-2024 for prototypes and a rollout production roadmap by the end of 2025.
An early access programme is currently underway, until January 2024, with a simple sign-up page on the website. The open release is scheduled for January 2024, and open APIs will be accessible to all developers by February 2024. Ola plans to integrate Krutrim across its group companies for sales, service, support, and other operational processes.
Krutrim: The Ethical AI?
The startup has been launched at a time when cases of misinformation and deepfakes are on the rise, with GenAI at the centre of the ruckus.
Aggarwal emphasised that the current safety debates around AI are heavily influenced by Western concepts. While Krutrim is being designed with safety in mind, he asserted that “the approach to safety will be grounded in an Indian perspective.”
The table below provides a comparative overview of how ChatGPT and Krutrim approach ethical considerations and biases. However, before that, it is worth noting that Krutim is yet to be launched in beta mode while ChatGPT is fully functional. Also, we are using the specs of ChatGPT 3.5.
Meanwhile, life seems to have come full circle for the homegrown GenAI ecosystem. Within months of Altman’s controversial ‘don’t even try’ statement, the country appears to be at the cusp of an AI revolution but the path ahead appears to be a long one.
Having joined the GenAI fray a little late, Krutrim will now have to spar with domestic players like Bharat GPT by CoRover.ai, Pragna by Soket Labs, Tech Mahindra-backed Project Indus and Sarvam AI in India and foreign entities such as OpenAI’s ChatGPT, Google’s Bard and Elon Musk-backed Grok AI to secure its market share.
According to Inc42’s “India’s Generative AI Startup Landscape, 2023” report, the country’s GenAI market is projected to surpass $17 Bn by 2030 from $1.1 Bn in 2023, growing at a CAGR of 48%.
With Krutrim, too, in the GenAI race, it remains to be seen who will emerge as the Indian AI juggernaut in the years to come, outpacing its Western counterparts.
The post Can Bhavesh Aggarwal’s Krutrim Take On ChatGPT With Its GenAI For India Playbook? appeared first on Inc42 Media.
login
who are we?
contact us
qosheapp
Toi Staff
Gideon Levy
Belen Fernandez
Mort Laitner
Rami G Khouri
Ali Fathollah-Nejad
Nikkei Editorial
visit website
