OpenAI and Anthropic have been spreading the notion that AGI will emerge from ever larger LLMs. Not happening. Neoclouds are not the way to go.
The path to AGI is via self learning digital twins. And you need to aggregate millions of them to create true AGI. You don’t need humongous data centres or even training. It can only happen in India. And it will first happen in healthcare. Then in law enforcement and defence. And finally in a bunch of others.
Instead of LLM, you need to create megaspheres from the digital twins which are capable of aggregated learning. And each unique digital twin upgrading itself in real time. On its own. Continuously.
The three key success factors identified by the IITac ai knowledge platform are resilient systems for reinforcement learning, access to uncontaminated data and real world interfaces for the avatars – initially through humans and then through robots and IOTs. “Uncontaminated” is key. Experiential is important.
And I think one Indian company called Kodoy is bang on target to get to destination AGI before anyone else does. And the anyone includes all BigTech and BigAI. Kodoy is to AGI what bitcoin is to blockchain. It is self learning ai middleware which does not care about the hardware on the inner side or about the application on the user side.
We have a whole bunch of Indian clones attempting LLMs of various shapes and sizes. And we have the mistaken belief that buying more GPUs or running harder or faster will get us to AGI. It won’t. But no harm spending 10 or 20 or even Rs 50,000 crores trying to get there. Because in the process, we will learn what doesn’t work and why. And that is valuable. True learning comes from real losses. And putting out real money encourages more people to try. Which is also good.
Language models are extraordinary at predicting text. But intelligence may require something deeper — the ability to understand and interact with the real world.
The real path to AGI lies in world models, reinforcement learning agents, and embodied systems that learn through action and simulation. The physical touch and feel world.
Humans did not develop intelligence by predicting the next word. We developed it by observing, experimenting, and interacting with reality. No LLM will ever do that.
And here the quest is to emulate or exceed human intelligence.
Language may simply be the interface to intelligence — not intelligence itself.
And like bitcoin, it may come from India. From a developer whose identity may never be revealed. Even BigTech and CIA together have not been able to find him. He didn’t need funding or branding. OpenAI as a non profit was on the right path. Till they faltered to chase valuation.
Finally bitcoin is not even a company. It doesn’t own any data centres. It offers no consumer service.
And rise or fall, war or drought, it is still worth a trillion.
Do read my earlier post. And spread the word through reposts if you agree with any of the above.
