Can Indian Startups Thrive Without Incentives In Making LLMs?

The crucial question in the current context is: Where support from the Indian Government is crucial for the development of Indian LLMs, and whether Indian startups in the AI space can realistically develop LLMs without government support.

This is further a broad reflection on the topic of whether government support is mandatory for large-scale AI development.

Let us take the global precedent to reflect upon the ground-level reality.

In reality, Global Large Language Models more or less across the world like GPT-4o, Gemini, Llama, Claude, Mistral, etc, have all been primarily invested in and developed by private enterprises and or startups with or without VC funding.

In most cases, there has been a large tech company like Google or a well-funded startup like OpenAI behind the development and deployment of such large language models. For example, OpenAI has a significant contribution from Microsoft, and Llama has a significant open-source ecosystem alongside a root contribution from Meta.

The above found reality is a reflection of the stand, augmenting the assertion that it is possible to develop resourceful LLMs even without government support. 

That said, if there is a sovereign AI requirement from a government perspective to offer cost-effective LLM resources for usage by citizens at large, it may be a strong recommendation for the relevant government to provide support for the infrastructure, like GPU support, to develop such sovereign LLMs. 

So, in fact, what I endorse is a middle stand, whereby the ecosystem should develop a healthy combination of sovereign models and commercial models.

That said, I strongly feel that the overall support needed for a strong startup ecosystem to build LLMs is not restricted to providing hardware support. It involves a myriad set of requirements (beyond the large hardware infrastructure support like GPU bank) ranging from –

• The need for a strong ecosystem of AI research
• The need for a set of well-curated localised data repositories
• A strong AI governance mechanism to oversee, monitor, and improve the models
• Last but not least — A strong human resource ecosystem to generate manpower right from data labelers, to data preprocessors, to data engineers, to ML algorithm specialists, to ML deployment engineers, AI policy specialists, and ultimately AI managers and chief AI officers. 

We believe we have a strong emerging ecosystem of AI research in India, comprising research from IITs, IIITs, NITs, and commercial research labs like Google Research India, Microsoft Research India, and IBM Research India. India ranks third in the world in terms of high-quality research publications in AI but is at a significant distance from world leader China, according to an analysis by research agency Itihaasa.

While it may lag the USA and China, we are seeing green shoots of a strongly evolving AI research ecosystem of academic and industry research. 

Regarding the data repositories, there is a significant contribution the government can make via its well-organised effort at AIKosh, the repository for Indian context data sets. 

This effort is very well appreciated and will be a game changer for the LLM development for AI startups and other ecosystem players. So, a significant effort in expanding the scope, variety, and veracity of the datasets in AIKosh will go a long way in streamlining Indian LLM efforts. 

Regarding emerging AI governance efforts, the IndiaAI mission under MietY (Government of India) is very soon going to launch an India-specific framework for AI governance, primarily focused on safe and responsible AI, while not stifling innovation.

Multiple MietY round tables have happened with stakeholders like academia, government, and startups, all to shape the same, guided by existing frameworks from UNESCO and the EU AI Act. 

On skills, I think India has the potential to be the skill powerhouse of the globe. This includes outputs from large AI/ML degree programs in engineering and management Institutes, AI/ML certification programs for continuing education among professionals, and to top it all, focused CXO awareness programs on AI/ML. These programs help India to emerge as the AI skill capital of the world. 

On these points in addition to the infrastructure angle, there is a definite view that government support can help a lot in accelerating the AI and LLM innovations. 

A multi-criteria support on all the above four fronts of data repositories, AI governance, AI research, and AI skill development, in addition to the availability of infrastructure, will be crucial for the development of India’s LLMs. 

These LLMs are a basic need of the hour, given the proliferation of LLMs in our daily lives and the need to contextualise the same for India-specific data and use cases specific to India.