- विज्ञापन -
Globally there is a ongoing race in the AI based supremacy between US and China. There are two major players in this race. The first is the United States, as united states and its companies like Open AI and Meta have gained dominance in the field of artificial intelligent.
The United States built foundational model called as large language models and these models have enable the companies to make transformative gains in the field of artificial intelligent. However, China and it’s companies like Deep Seek as well as Ali Baba, they are trying to make their own foundational models or large language models which according to them have been made at much lesser cost than the Americans. But in this india lacks currently quite far behind.
Should India use its limited resources to develop a foundational model like large language models on it’s own or should it depend on the already existing western models?
For the understanding we have to first understand the research and the kind of efforts countries are putting in the fields of AI. USA is currently the most dominant player and it is trying to maintain it’s dominance by undertaking a project called as Stargate as the stargate project which involves about 500 billion dollars in funding which will be provided by major stakeholders in the field of AI that is Open AI, Nvidia, Microsoft as well as Elon musk lead platform X, these models or these companies will pitch in for the 500 billion dollar funding which will enable USA to create an AI based infrastructure which can be used in years to come.
On the the hand, there are companies like Deep Seek which has brought its own foundational level model and according to Deep Seek, its foundational model or the LLM has the ability to match the capabilities of Open AI’s own model.
This is the race which has been currently involved between countries like United States and China.
However, there is a debate going in India whether or not India should develop its own large language models.
Now to simplify, what is the large language or a foundational model for AIs is that these models use large amount of data whether its voice text or image based and by using these data these large language models understand the context, the grammar, the patterns and the nuances that are associated with human languages and when these data are processed using the graphic processing units or the GPUs it helps the AI model to understand and mimic human languages and accordingly respond to the commands of the user based. The most important aspect of any AI model are two things; first it need large amount of data and to process this data its require a large amount graphic processing units. This is the challenge that is associated with the development of a new foundational that it requires large amount of data and it requires large amount of processing power that is with the supply of graphic processing units.
In this regard, Nandan Nilekani, co-founder of Infosys and who also played a very important role in the development of Aadhar based UID in India said rather than building or spending its resources building its own foundational model, India should rather rely on the models of the wester nations like US and using these models and its capabilities, India could apply it on a use case basis in Indian applications, He also added that it would save a lot amount of resources and also it will provide India with the speed with which it can utilize these foundational level data from US and accordingly innovate its own specific conditions .
While on the other side of this debate, Aravind Srinivas who is also the CEO of the startup called Perplexity AI said that India should indeed take every effort in order to make its own foundational level of LLM that is large language models as these LLM it could be based on Indian languages and culture. It could also enable India to get parity when it comes to benchmarking the capabilities of LLM is concerned.
If India were to take a development of its own LLM, it will require a large amount of graphic processing unit which will provide the ability to scale and undertake the large amount of processing. But currently the supply of these GPUs are a virtual monopoly of the country like Nvidia as most of the GPUs that are used in AI based processing and training models are sourced from just one company which is US based company.
To process large amount of data, require a GPUs and these can be quite expensive in nature so to enable Indian companies procure this GPUs and use it for their own AI based purposes the Indian government brought a scheme called India AI Mission as this scheme which provides funding of about 13370 crores, these funding will be used to procure about 10,000 GPUs and these GPUs then will be provided at a much subsidize cost to Indian companies that can use it for its own AI based purposes but every AI need GPU. However, in its last day of administration the previous US president Joe Biden, he brought regulation and this regulation, it favored restricted exports of GPUs.
United States closet allies which are 18 countries like Australia and Japan, they will have an unrestricted access to the US manufactured GPUs, whereas countries like China, North Korea and Russia which are adversaries of United States, they will be prohibited from the importing any US manufactured GPUs.
Moreover, there are two tiers; first tier consists of US closest ally whereas last tier consists of adversarial nations of the US that is China, Russia and North Korea whereas in between there is a second tier which consists of many countries including India, they will be subjected to a capped use of US based GPUs. India can import about 50,000 of GPUs until the year 2027.
If India were to take a large scale development of it’s own foundational model, it will require large amount of GPUs. However the supply of these GPUs as per the US regulations is limited to countries like India which poses resources base challenge for the India that even if it has money to import the GPUs or whether it will not be able to import GPUs is the matter of debate.
