the simple magic of prediction

I keep coming back to the idea that the most profound shifts in tech often hide behind the most deceptively simple math.

I recently watched a video on Markov chains ¹, and I’m finding it fascinating how much of our world is just a sophisticated version of “what happens next?”

Take the Markov chain itself. At its core, it’s just a “memoryless” process. The next state depends only on where you are right now. It feels like a toy, something you’d use to model a weather pattern or a simple board game. You have a state, a few transitions, and some probabilities. That’s it.

But when I see that simplicity scaled, the results start to look like magic.

A trillion dollar walk

I find it wild that PageRank, the algorithm that essentially built the modern web, isn’t just “inspired” by Markovian thinking. It is a Markov chain.

Early Google treated the entire web as a massive graph where a “random surfer” ² just clicks on links forever. Mathematically, PageRank is just the stationary distribution of that walk. It’s the probability that, if you clicked links at random for eternity, you’d end up on a specific page.

It’s a beautiful, yet absurd thought: the entire organization of human knowledge, distilled into a trillion-dollar company, was built on a “memoryless” walk through a forest of links. I’m realizing that the web wasn’t “indexed” so much as it was “sampled” into a hierarchy of importance.

Finding the primitives

I’m realizing that understanding these primitives is the ultimate multiplier in a post-LLM world.

For me, this is where the real work happens: in the sparring phase. Before I let an agent touch a file, we argue from first principles. But I have to find that first principle first.

I find it fascinating that we haven’t actually “solved” consciousness or reasoning. We’ve just found a way to make prediction so accurate that the distinction starts to fade.³

It makes me wonder what other “simple” concepts are waiting to be re-discovered and applied in a novel context. I’m excited to keep building. It’s not a database. It’s not a brain. It’s just a very, very good guesser, and that guess is currently re-writing the world.

Footnotes

1. Veritasium, “The Strange Math That Predicts (Almost) Anything”. A look at the history of Markov chains and their role in modern technology. ↩

2. Wikipedia, “PageRank”. The “random surfer” model treats users as memoryless agents clicking links. ↩

3. The nature of consciousness remains one of the great philosophical and scientific questions. Whether we’ve come close by simulating the outputs of reasoning, or if the simulation itself constitutes a form of reasoning, is a debate that’s only just beginning. ↩