India and its Own Large Language Model (LLM) — Do We Need One?
During his last visit to India, Sam Altman, CEO of OpenAI, effectively dismissed the idea of a Large Language Model (LLM) coming out of India. There had been pushback, mostly emotional but some were serious. I have however not seen any policy paper by either government or any private entity on this, on the merits and demerits, and a roadmap in case we are keen.
How many LLMs are there? Well, any estimate like this usually has an accuracy challenge. Wikipedia lists 33 of them, and quite a few are versions of the same model. And HiggingFace has 663 of them on the Open LLM Leaderboard (https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard).
Now we need to understand that while the large majority of these General Purpose models are Open Source, almost all the popular ones like GPTn (OpenAI), and PaLMn (Google) are proprietory. LLaMA of Meta is available for non-commercial research.
India does not figure in the world of LLMs, which is almost entirely dominated by the USA and China. This prima facie is not a concern, for all of them provide with Application Program Interface (API) to harness their power. Thousands of applications have been built on top of these APIs, including the fancied Jugalbadi (https://www.jugalbandi.ai/) using OpenAI GPT. But this does not answer the obvious questions:
1. With the China option being off the table in the foreseeable future, will putting the significant eggs in the US basket for a critical area like AI is a safe bet? Also, like any tech provider, how do we handle at least a generation lag in technology that we will have access to?
2. What about all the data that will be exposed to these models, including the critical data (for every commitment to data protection, we know about Cambridge Analytics)?
3. While all Indian IT Services companies are boldly announcing “platforms” “harnessing the power of AI” to help their clients, even a cursory look at the products, people, and publications AI space reveals no signs of such prowess. Given the lack of R&D and the potential conflict with clients on products (like Microsoft Google or Meta will be less than happy if they find a TCS or Infosys, or Cognizant is entering as a competitor). And I am not even talking about the patent disputes that will be thrown at if there is any credible progress.
4. And finally, if India is an IT superpower, should it not have an LLM for fair play and bargaining?
Unfortunately, there is hardly any open debate on where exactly India stands on capability and plans. Mostly we see either grandstanding by the corporates or silence from policymakers. But an honest discussion is needed, and urgently.
Let’s take a segway with an assumption, that we plan to build India LLM. For sure it can’t be done by the IT companies, it does not fit their business model. We will need a government initiative — much along the lines of India Stack. So if we plan, what do we need, and what do we have? Let’s look.
1. Purpose and leadership — well not visible at the moment. But if I just look at our space and nuclear history, there is no doubt that we can do it if we plan to do it.
2. Funds — how much it costs to build an LLM? No clear guidance. OpenAI investment is so far USD 11.3 Billion, but for sure it is not all burnt. A Google search says it will be at least USD 10 million. Most probably it will be tens of millions, not hundreds. So yes, if there is a will at the highest level, it can be funded.
A point to note here is though the Indian startup ecosystem is vibrant, almost 90% of the risk capital came from outside the country (a large part of it, to our great chagrin, from China). The reason is simple, the old money does not want the risk associated with tech, and tech new money is still a small pool (a tech billionaire in India pales in comparison to the Silicon Valley doyens). Whatever is being invested is in small parcels and on proven ideas.
3. Talent — a big question mark. Come what we say, we seriously lack talent. The best and the brightest (and they are at most in the hundreds) have been soaked by the GAFAM and precious little is left. And, for them, India is not hot, at least for AI. But miracles can happen with committed credible leadership, clear vision, and challenging goals.
4. Policies — I have concerns. And possibly I am right because people in power are not expressing urgency while the world is pressing the gas. Given that we believe in conservative rather than liberal governance, we are more concerned about the dangers than the possibilities. And that makes any attempt to build an LLM in India fraught with uncertainties.
I will be happy to be proved wrong on this count.
What do you think?
