Given the recent hype around generative AI in India, Jaspreet Bindra, founder of Tech Whisperer Ltd., along with Sudhir Tiwari, Managing Director at Thoughtworks, have proposed that India should build an equivalent to ChatGPT called JanAI or जन AI, or AI for the people.

In a recent piece, “our proposition, however audacious it may seem, is that India should consider building generative AI as a digital public good – we call this JanAI or GenAI for the people.”

Bindra and Tiwari propose that for this to take shape, India needs to take a collaborative approach and serve generative AI as digital public good, much like Aadhar and UPI. “We believe that this could happen through a tripartite partnership between a proactive government, our world-leading IT industry, and some leading technical institutions like the IITs.”

It isn’t for the lack of talent or government support and India has the capability to build local LLM-fine-tuned with data for Indian languages they say, but, “We believe, however, that it is more important to consider the objective and aim for such an exercise. Should India follow the western capitalist method, or the Chinese State-controlled one?”

In March, CoRover.ai announced BharatGPT which is yet to be launched. The model supports 12 Indian Languages while also hosting more than 120 foreign languages. The company offers text, audio and even video. Similarly, Bindra said that India should further fine-tune its own India-contextual LLM, calling it BharatLLM.

In July last year, The Ministry of Electronics and Information Technology (MeitY) within Digital India Corporation launched Bhashini, a repository of digital content in all Indian languages. Universal Language Contribution API (ULCA) now has the largest repository of Indian languages. 

Harnessing the API, Tech Mahindra also announced an Indic-based foundational model named ‘Project Indus’. The language model will support 40 different Hindi dialects with a plan to add more languages subsequently. 

While the others are ahead of the race, India is still figuring out how to gather datasets of Indic dialects to train LLMs in our language with Project Indus relying on ‘bhasha daan’.

Meanwhile, other countries are working towards ethical and responsible development like the EU, and UK. Last week, researchers from UAE in collaboration with Cerebras introduced the Arabic alternative to ChatGPT, Jais is a 13 billion parameter model based on 116 billion Arabic tokens trained on Arabic websites, books, news and Wikipedia.