Sign up for our daily and weekly newsletters to stay updated with the latest industry-leading AI coverage and exclusive content.
When Sam Altman visited India last year, he remarked that it would be impossible for a startup to compete with OpenAI at training foundation models with just $10 million. This statement grabbed headlines, and CP Gurnani, the former CEO of Tech Mahindra, responded by accepting the challenge to build generative AI in India.
Fast forward to early 2024, and India, known for its tech talent, is making strides in generative AI. Surprisingly, the first Indian entity to make a significant move against OpenAI’s GPT models isn’t Tech Mahindra but a startup founded by Bhavish Aggarwal, who also founded Ola Cabs. Ola Krutrim, meaning “artificial,” launched its first language model, Krutrim base, along with a chatbot last month and plans to go mainstream soon. Other players like Tech Mahindra and Reliance Industries are also in the race, trying to catch up.
Localized Experiences
While foundation models like OpenAI’s GPT and Meta’s Llama are proficient at generating language, answers, and code, they often struggle with non-English languages, especially low-resource ones. To address this and enhance localized experiences, companies in South Korea, Finland, China, and now India are developing proprietary models to increase local language and cultural representation.
India’s generative AI challenge is substantial, given its 1.4 billion population, 22 officially recognized languages, 1,600+ dialects, and 19,200 unofficial dialects. Training a model to encompass all of this is both challenging and capital-intensive.
Aggarwal founded Krutrim in April 2023 to tackle this issue, raising $24 million from Matrix Partners and training the model with two trillion tokens, the largest representation of Indic languages to date. Currently, Krutrim understands 20 Indian languages and generates text in 10, including Hindi and English. According to the company, its performance in Indic languages already surpasses GPT-4, though its English capabilities lag behind but are expected to improve soon.
Krutrim is developing in phases, with plans for support across all officially recognized Indic languages and a Pro version for complex problem-solving, including text, vision, and speech. They have also built a ChatGPT-like chatbot for the Indian market, though it’s not publicly available yet. Additionally, the company is researching hardware to build its AI supercomputer.
Competing Players
Krutrim is among the first Indian companies to cover all bases in generative AI, but other notable firms like Tech Mahindra and Reliance Industries are also making strides. Tech Mahindra, under CP Gurnani’s leadership, launched The Indus Project in August 2023, developing an open-source large language model (LLM) for Hindi, which is set to debut in February 2024 with 539 million parameters and 10 billion Hindi and dialect tokens.
Reliance Industries, known for leading India’s 4G wave with Jio, announced plans to build language models for India at its AGM last year. Partnering with Nvidia for AI infrastructure, Reliance is developing Bharat GPT with IIT-Bombay. While details are sparse, it seems Reliance aims to integrate GPT across its customer-facing products and services.
Bengaluru-based Sarvam AI, backed with $41 million in funding, recently emerged with a 7 billion parameter Indic language model based on Llama2, planning an enterprise platform for generative AI apps. Google-backed Corover also claims to support 22 Indic languages for its conversational enterprise chatbot platform.
Enhancing Generative AI Experiences
As the ecosystem evolves with more players and advanced technology, India expects more sophisticated closed and open-source Indic language models. These developments will not only enhance internal enterprise workflows but also improve applications across various sectors. For instance, Tech Mahindra envisions its Indus Project LLM helping over 140 million farmers with information on loans, pesticides, and other agricultural needs in their preferred languages. It could also power healthcare and finance kiosks to interpret local dialects and provide instant information, showing the endless possibilities of generative AI.
It will be fascinating to see how these models compare with global counterparts like OpenAI’s upcoming GPT-4.5 and Google’s newly launched Gemini series.