Within days, another Chinese company announced a model that could process 20 to 32 times more inputs than any company in the world today. This equates to processing a 16,000 page PDF file.
These developments pose sobering questions to us. Why isn’t this kind of innovation happening in India?
Also read: India needs dedicated AI cadre to effectively govern this technology
The AI revolution is unfolding before our eyes, and a few quarters ago we were all on the same starting line in model development.
In theory, we were well placed to lead it.
The country boasts a large number of AI engineers and programmers, a thriving technology ecosystem, and a burgeoning pool of risk capital.
There were no historical handicaps in model development (as there were in chip manufacturing).
This formula was known not only to the United States, but also to model makers in China, Korea, Europe, and West Asia.
However, the generative AI landscape in India remains largely derivative.
While some companies are tweaking open source models for Indian languages and specific applications, there is a worrying lack of fundamental breakthroughs similar to GPT-4, Claude, and DeepSeek. There seem to be very few companies.
This raises serious questions about our approach not only to GenAI model development, but to all types of basic research.
Missing foundation: When the transformer model first emerged as an AI concept a few years ago, India had the talent and resources to tackle basic research.
However, the focus remained primarily on the application and adaptation of existing models.
Also read: AI needs to be human-centered to attract people and lead to adoption
The general view was that building models from scratch was too expensive.
We chose to skip the competition rather than address the cost challenge of affordability.
This myopia is troublesome.
The core principles of the transformer model are neither esoteric nor inaccessible. They can be explained through basic mathematical operations.
But rather than leveraging local talent and exploring new architectures and more efficient ways of doing things, we chose to build on another company’s foundation.
Currently, the number of research papers by our experts is only one-tenth of the number of research papers in China or the United States.
As a result, India will continue to pay for intellectual property in perpetuity, including cloud usage fees, licensing fees, and the cost of foreign hardware.
At some point, participating in the global AI race will be as difficult as it is currently in cutting-edge chip manufacturing.
The cost of incrementalism and over-focus on the end goal: The global AI race is young and there are no clear winners.
China’s rapid progress shows what is possible. Its LLM is not only efficient but also versatile. Impressively, these models seamlessly support multiple languages, including Indian languages.
In contrast, the most laudable Indian efforts have focused primarily on adapting existing basic models to support local languages, a strategy that addresses immediate needs but that We have not created products that could not only foster innovation but also strengthen our global competitiveness.
Without fundamental breakthroughs, even India’s indigenous language models risk being overshadowed by foreign alternative language models that offer similar functionality at scale, in addition to a vast array of other features. .
Innovation centers around the world are investing heavily in basic AI research, large datasets, and in-house training.
Our application-oriented thinking is rooted in a broader problem: a lack of appetite for funding research without a clear end goal.
Also read: IT ministry’s AI regulatory report widely welcomed, but not entirely
Basic research is speculative in nature.
It requires trial and error, and there is no guarantee of success. The situation is even worse when developing AI models.
You may not fully realize the benefits of formula adjustments and other architectural experiments until the model is fully trained.
However, the rewards of success can be transformative. Countries that invest in such research today will shape the industries of tomorrow.
Dependency is expensive: History teaches us a harsh lesson: Dependence on critical raw materials and technologies can set a nation back further.
Failure to develop basic capabilities risks creating paralyzing dependence on foreign technology in areas of high innovation where it is impossible to catch up with leaders.
As global AI capabilities expand, the cost of leveraging other companies’ innovations will become even higher. Without ownership control of the base model, adapting it to your own needs can also be hampered.
GenAI, like software, has the potential to be a significant source of net foreign exchange revenue for us, but at today’s pace it risks becoming a huge foreign exchange absorber like oil.
Dependency on India extends beyond GenAI.
Robotics, self-driving cars, AI-enabled drug development, and other emerging industries are also taking shape.
Fundamental research is still needed in these areas, and opportunities for leadership remain.
But if India remains in its old ways and relies on incremental adaptation rather than undertaking bold exploratory research, it could also fall behind in these industries.
We need to pivot quickly. Transformative breakthroughs often emerge from uncertain beginnings.
As long as it involves bold experiments or basic research, efforts without a clear end goal require financial support.
India’s future in AI and many other emerging technologies depends on visionary leaders who are willing to invest in the unknown.
The author is an innovation investor at LC GenInnov Fund, based in Singapore.
