Ola founder Bhavish Aggarwal on Friday (December 15) unveiled AI models of his new startup Krutrim AI and touted its solutions as “India’s first full-stack AI”.
During the launch event, Aggarwal also showcased an AI chatbot, powered by Krutrim, which functions in a way similar to other open source large language models (LLM) such as OpenAI’s ChatGPT, Meta’s Llama 2, and Google’s Bard.
“Today, all AI models, called LLM, are trained largely in English. But language is just not text but also the vehicle for cultural values, context, and ethos. Hence, the current AI models just can’t capture India’s culture, knowledge, and aspirations, given our multicultural, multilingual context. An India-first AI… needs to be trained on datasets specific to us,” said Aggarwal.
Krutrim AI claims to have built its AI models from scratch, having trained them on 2 Tn tokens and unique datasets. The AI models can understand over 20 Indian languages and generate texts in 10 Indian languages, including Bengali, Tamil, Malayalam, Gujarati, and Marathi.
The word Krutrim is taken from the Sanskrit language and means “artificial”.
The sign-ups for the base model ‘Krutrim’ will open today on the Krutrim website. Its next model called ‘Krutrim Pro’, with more sophisticated problem solving capabilities in text, speech, and vision, is expected to be launched in the next quarter.
Aggarwal incorporated the startup, Krutrim Si Designs Private Limited, in April this year. Recently, the startup also raised $24 Mn in debt funding from Matrix Partners.
The post Bhavish Aggarwal’s Krutrim Unveils ‘AI Models For India’ appeared first on Inc42 Media.
