how can you not be romantic about LLMs?

In Fossil Future, Alex Epstein describes how millions of years of organic waste, immense pressure, and time sedimented beneath the earth to become fossil fuels—concentrated, volatile energy that now power modern civilization. This epic, geological process is how I’ve been thinking about the rise of Large Language Models. And it makes you wonder: how can we not be romantic about that?

For decades, we have been pouring ourselves into the internet. Questions. Answers. Arguments. Dreams. Doubts. Jokes. Music. Research. Pirated books. Documentation. Mistakes. Breakthroughs.

Every Google search. Every late-night curiosity. Every tutorial someone wrote to understand something just a little better. Every forum fight. Every stack of documentation. Every thread we abandoned halfway.

All of it: billions of tiny data points scattered across the digital universe. Our queries and curiosities, a vast, continuous syntax of learning nodes, endlessly connected like cordyceps.

Artificial Intelligence Model Life Cycle: From Creation to End-users | Gcore

At some point, it would’ve been absurd if all of that didn’t turn into something. If it didn’t recycle, compost, and recombine into a new form of intelligence—the standard by which we now define LLMs. The direct result of humanity documenting itself obsessively for 30 years, or what happens when collective memory becomes machine readable. When human behavior becomes data. When questions become training sets, and the internet finally learns to speak back.

LLMs are one of the most human things we’ve ever built. They are built out of us: our language, our curiosity, our contradictions.

So yes, it makes perfect sense that AI is here. In fact, it would’ve been strange if we never built it. We’ve been preparing the ground for decades. And now the soil in our digital landscape has sprouted something powerful, strange, useful, urgent, and undeniably beautiful.

How can you not be romantic about that?