Open ai and Anthropic are like Android and iOS. But these guys have both got it wrong. LLMs are about selling USD 100 notes for USD 1 and reporting revenues.
Android and iOS both purchased the LLVM from UIUC and called it the “app”. Then they made tweaks to differentiate. But they were co-operating competitors who killed the rest. And owned the world. Till ai came to eat their lunch.
App has been replaced by LLM.
But apps cost very little to host.
And actually made money.
This will never be the case with LLM.
Let me start with voice ai riding on LLM. It costs Rs 20 per minute to converse with OpenAI or Eleven Labs. Sarvam claims they will do it for a rupee. Depending on how much you are willing to lose, you can even do it for free. No matter what you do, compute will be at least Rs 100 per hour and that sets the base price.
So this entire LLM business and services like voice or document reading which ride on it are a competition for losing money. Ai has potential for drastic cost reduction, but LLM does not. For Ai to be effective, it has to be played differently. And to be effective and cost efficient, it cannot run off a central data centre or be developed using scant or synthetic data.
Actually it can’t even be trained if it has to compete with the human brain. As, we go from AI to AGI (artificial general intelligence) – it is no longer about some overfunded copycats competing with each other. The battle field shifts to competing with the human brain.
And no one has a clue on how to do that. And because, we don’t know how the human brain works – how can we clone it.
LLM presents a simplified view. Collect some data. And train the ai with the data. It is like teaching a dog to play with a ball. Or teaching a parrot to repeat. And after this so called ai has been trained – humongous amounts of compute are required to make it work like a trained dog or parrot. It is entertaining and has huge novelty value. If your job is to fetch a ball back. Sure the trained dog can do it for you. And put you out of work. Or if your job is to repeat what your boss says like a parrot – the trained parrot is serious competition to your livelihood. So if you do a repetitive task or a “hear and repeat” role – you are now obsolete.
But that is just not radical enough. Not as yet at least. It becomes radical only if the ai can learn. It has very limited value if it can only do what it has been trained to do.
Humans or most animals learn on their own. A baby elephant doesn’t need to be trained in a school to learn how to walk. Nor does a cat go to college to learn how to catch a rat. They are not taught. They learn on their own – at least what they need to survive. Dogs were not designed to fetch balls nor were parrots designed to emulate humans. So that has to be taught.
OpenAI and their lookalike Anthropic are both wrong in believing that LLM is going to scale. It won’t.
Read my next post on what will work, if you agree LLM won’t.
