LLMs

I’ve been wrestling with large language models at work for about 2 years now, to get them to do the things that I want. And what I’ve come to realise is that we might have accidentally stumbled upon something truly groundbreaking here, and still be sleeping on it.

I use python as the interface language to communicate with these models, and there is this sense of profound excitement whenever I run my little py file to get an output. “Does it understand what I’m asking? How good is it at doing that? How do I align it to get the stuff that I want?” Lately, I’ve been hazily dreaming of prompts, which is a little extreme, but I think it’s a good trade-off. The most incredible thing about an LLM is that it understands human language exceptionally well.

I am witnessing the birth of two powerful unlocks here.

Language, as we know it, is a powerful wand. It is a medium to convert intent, to invoke an action and comprehend feedback. This is agentic behaviour, and if you have a personal machine doing that, you have a self-functioning entity that is an extension of you. You are no longer just a lone, powerless individual — you are something more. With this unlock, there is a deep well to explore here, and I think we have only started to scratch the surface.

The second unlock is something I’m truly excited about. If you can get a model to understand your language, in theory, you could command or instruct it to reason about ideas. If you can get it to reason about ideas, no matter how unagentic it might be, you still get something extraordinary for free — compressed intelligence conversing with you in perpetuity.

I am of the belief that having dialogues / conversations is the path to think better. With that possibility here, you can now truly focus on the task you have been working on all your life — to push your brain to limits it has never seen before. 10x intelligible or even more.

Leveraging a new technology to become smarter might sound like a tagline from some cheap self-help book. But there is a lot to be gained here. Being a smarter individual consists of having a good understanding of a variety of fields, being able to build a concept from first principles, making better decisions by not falling into fallacies, asking better questions, writing better, speaking better and getting into the habit of consuming new knowledge effectively — which is essentially understanding how to learn. These are foundational skills, and improving on them gives unimaginable benefits across everything that you do.

LLMs are able to help in every one of these atomic skills if you know how to work them.

The ones who will win are those who will put in effort into intellectually evolving with this free aid — not merely using it to get work done.

Machines have always been an extension of our physical capabilities — cars, tools, heck even a tiny screwdriver. They all served a purpose for us, something that we were inherently not able to do with our physical bodies. We now have something that exists as a catalyst to push the frontier of our intelligence.

We are at the beginning of the age of self.

Cooper's Farm, Interstellar

I’ve always been awed by this scene from the movie. The tractors are autonomous entities, but when their programming went haywire, they instinctively returned to their master—Cooper—so he could repair them and send them back to work. In this moment, they felt like an extension of him — machines he was orchestrating. Later, he would go on to command even more advanced ones — TARS and CASE.