In a bid to revolutionize the Artificial Intelligence (AI) landscape, Ola co-founder Bhavish Aggarwal’s startup, Krutrim AI, has launched its first family of multilingual AI models. With the aim of addressing the inadequacies of existing models in capturing India’s diverse cultural and linguistic nuances, Krutrim AI introduces a base model and a more advanced Krutrim Pro, marking a significant milestone in the country’s AI development.
Krutrim AI’s Inception
Founded in April 2023, Krutrim Si Designs Private Limited emerged from the visionary collaboration between Bhavish Aggarwal and Krishnamurthy Venugopala Tenneti, a prominent board member of ANI Technologies Ltd, the parent company of Ola Cabs and Ola Electric. Aggarwal’s determination to create “India’s first full-stack AI” led to the establishment of Krutrim AI, aiming to break free from the English-centric model training prevalent in the AI industry.
The term “Krutrim,” derived from Sanskrit, meaning “artificial,” signifies the essence of these groundbreaking models. Krutrim AI sets itself apart with two distinct models – Krutrim and Krutrim Pro. The base model, trained on an impressive 2 trillion tokens and unique datasets, is a testament to Krutrim’s commitment to capturing the rich tapestry of Indian languages and cultures. The larger, more intricate Krutrim Pro, scheduled for an early launch next year, promises advanced problem-solving and task execution capabilities.
Unveiling Krutrim’s Potential
During the launch event, Bhavish Aggarwal provided a glimpse into Krutrim’s potential by showcasing an AI chatbot powered by the base model. Comparable to established models like OpenAI’s ChatGPT and Google’s Bard, this chatbot boasts an understanding of 22 Indian languages and the ability to generate text in 10 Indian languages. This showcase reflects Krutrim AI’s determination to offer solutions tailored to India’s multicultural and multilingual context.
Challenges to Existing Models
Aggarwal highlighted a critical limitation in existing large language models (LLMs), stating that they are predominantly trained in English. This, according to him, results in a failure to capture the essence of India’s diverse cultural values, context, and ethos. Krutrim AI’s emphasis on training its models on datasets specific to India aims to overcome this challenge, positioning itself as an “India-first AI.”
Building from Scratch
Krutrim AI claims to have built its AI models from the ground up, starting with a comprehensive training regimen involving 2 trillion tokens and unique datasets. This approach reflects the startup’s commitment to ensuring its models encapsulate the richness of India’s culture, knowledge, and aspirations. The ability to understand over 20 Indian languages and generate text in 10 languages, including Bengali, Tamil, Malayalam, Gujarati, and Marathi, showcases the inclusivity of Krutrim AI.
Signifying a pivotal moment in India’s AI journey, Krutrim AI opens sign-ups for its base model, ‘Krutrim,’ on its website. Additionally, the anticipation builds for the imminent launch of ‘Krutrim Pro,’ set to bring even more sophisticated problem-solving capabilities in text, speech, and vision in the next quarter.
Recently, the company secured $24 million in debt funding from Matrix Partners, adding a significant financial boost to its endeavors. This funding signals the confidence investors have in Krutrim AI’s potential to disrupt the AI landscape and pave the way for a more culturally sensitive and linguistically diverse approach.