DeepSeek and Geopolitics of LLMs: Avenues for South Asian Cooperation Notes for India and Nepal

Kathmandu: Large Language Models (LLMs) are the bedrock of any artificial intelligence (AI) system. Through these models, a computer is taught a more humane intelligence, which enables increased and diversified interactions between machines and humans. LLMs have immense benefits in different arenas of governance, be it health, education, transport, defense, or security.

Nevertheless, creating LLMs requires massive funding. It also includes strategic equipment such as semiconductors for development, which complexes the processes for making LLMs. Yet, the race for LLMs kick-started with the USA’s launching of LLMs of OpenAI, Google’s Bard, Meta’s AI, and Microsoft’s Copilot. Already rich tech giants of Silicon Valley, supported these ventures, with huge investments.

Nonetheless, a prominent disruption in the LLM market has been created by the Chinese start-up DeepSeek. The researchers behind the innovation claimed that it has been made at a relatively lower cost than any other AI model. The BBC cited the researchers as claiming the application came into operation for $6m to train, a fraction of the “over $100m” alluded to by OpenAI boss Sam Altman when discussing GPT-4.

DeepSeek’s creation of an affordable, open-source LLM has subverted Western tech behemoths’ hegemony and caused a reassessment of artificial intelligence research policies around the globe. This change especially affects smaller countries like Nepal and offers special chances for nations like India, with its varied collection of Indo-European languages, to create native LLMs that would help the larger South Asian area. There are great geopolitical consequences.

The above graph shows the exponential growth in the Large Language Models (LLMs) in the past five years. 

DeepSeek’s achievement points to the fact that the AI race is not limited to nations with significant technology and financial capability. Smaller countries might be able to create their artificial intelligence capacity thanks to this democratization, hence lessening reliance on world tech giants and promoting technical sovereignty. But it also begs questions about the possibility of more monitoring and control in non-democratic regimes. DeepSeek is not free in the manner in which it analyses data, unlike AI models created in the West.

DeepSeek is designed to avoid discussing politically delicate subjects, given China’s tight internet rules and content control measures. Inquiries about events like the 1989 Tiananmen Square demonstrations, China’s human rights policy, or problems of sovereignty between Taiwan and Hong Kong are either greeted with evasive, non-committal responses or outright prohibited. This degree of control goes beyond just political subjects. DeepSeek ensures that material contradicting state-approved narratives stays clean by filtering materials on democracy, human rights campaigns, or specific global conflicts.

Oli’s Chinese Detour: Implications on Domestic and Geopolitical Landscape

Furthermore, China’s AI models often provide a perspective that fits its geopolitical goals, which shapes user behavior—particularly that of countries without their artificial intelligence capacity. With 22 officially recognized languages and several dialects, most of which fall into the Indo-European language family, India’s linguistic scene is marked by its great variety.

In the framework of artificial intelligence development, this variety offers both a possibility and a difficulty. Creating indigenous LLMs specifically for India’s linguistic pluralism guarantees that AI uses are relevant culturally and easily available to a larger populace. By offering information in several languages, such models may help public services, educational resources, and effective communication to be improved.

Moreover, indigenous LLMs help to conserve language legacy and advance digital space inclusiveness. Geopolitically, India’s investment in its LLMs may help to lower reliance on outside technology, thereby improving national security and technical sovereignty. It also makes India a leader in artificial intelligence within the South Asian area, allowing it to help other nations like Nepal acquire their AI capacity. In the digital economy, this cooperative strategy may promote regional integration and common progress.

Particularly in areas like linguistic representation, digital accessibility, economic growth, and technical independence, India’s creation of a large language model (LLM) specifically for South Asia might provide Nepal with major benefits. Access to a model that fits Nepal’s language and cultural demands will be helpful as artificial intelligence becomes more and more included in government, education, industry, and everyday life. Many global artificial intelligence models have great difficulty supporting smaller languages.

Although Chinese models give Mandarin and regional Chinese dialects top priority, models established in the West mostly concentrate on English and other widely spoken global languages. Given its unique multilingual environment, an AI model created in India would probably include Nepali, Maithili, Newar, and other languages, hence increasing the availability of digital tools to a larger population of Nepal.

Such a paradigm might raise the contextual relevance and accuracy of voice recognition, machine translation, and content creation driven by artificial intelligence. Businesses, media, and government institutions in Nepal might employ artificial intelligence, for example, for improved native language communication, hence lessening dependence on English-based solutions that might not completely reflect local subtleties.

DeepSeek marks a new era in the worldwide artificial intelligence (AI) arms race, where reasonably priced, open-source models are no longer exclusively under the control of Western IT companies. DeepSeek’s built-in filtering systems, however, pose questions about digital sovereignty, especially for developing nations like Nepal.

Investing in indigenous artificial intelligence models from India offers an effective replacement that gives language inclusiveness, freedom of information, and democratic principles priority. Developing its LLMs would help India not only improve its own AI ecosystem but also provide its neighbors in South Asia with tools that represent the political and cultural variety of the area.

South Asian countries need to work together to create autonomous LLMs that respect their own language and governance systems, therefore guaranteeing a fair and open AI scene. Dependency on censored artificial intelligence models such as DeepSeek runs the danger of restricting access to objective knowledge, therefore strengthening foreign digital dependencies and compromising regional technological sovereignty. Rather, nations like Nepal and India can build a more open, inclusive, and representative digital future by funding locally built artificial intelligence technologies.

(Harsh Pandey is a PhD Candidate at the School of International Studies, JNU, New Delhi)