01The Slop Trilogy · part three

He counted letters in a poem.

In 1913 Andrey Markov went through Eugene Onegin by hand to prove something about probability. What he ended up with was the idea that you can predict the next thing from the current thing without understanding either.

02Its most honest descendant

It doesn’t know what you’re saying. It knows what usually comes next.

For about a decade that was the whole of it — a small convenience with an obvious ceiling, because a chain that only knows the last word or two produces nonsense almost immediately. Then two things happened at once.

03Two lines, separate for a century

Then not

Neither line was built to do what the pair of them now do together.

1913Markov counts letters in a poem. Predict the next thing from the current thing, without understanding either.

1994The cookie is invented so a shopping basket survives a page refresh. A small note the site keeps about you.

2007Predictive text ships on phones. The chain in your pocket, guessing the next word.

2010sThe note gets long. Cookies become profiles: what you read, bought, hovered over, abandoned.

2020The chain gets a memory. Neural networks let it condition on thousands of words instead of two.

NowThey meet. A vendor keeps a profile of you and a predictor writes with it. The note is no longer read — it’s used as the prompt.

04Not about capability

What changed is what it’s predicting.

For a hundred years the chain predicted text. Letters, then words, then sentences. Now the same mechanism — still predicting the most likely next thing, still with no idea what any of it means — is being pointed at the output of your professional life.

05Interactive · the chain

Same seed, two corpora

An order-2 chain. Pick an opening and it predicts its way forward from what it has read.

Pick an opening.

Build it from your file instead

A real chain over two small corpora. Untick the box and it has read generic business prose. Tick it and it has read the same prose plus a context file. Nothing is sent anywhere.

06What the box changes

Not because it got cleverer.

Untick it and the chain walks toward the middle of everything it has read, because that’s where the probability is. Tick it and the same seed goes somewhere specific — because the distribution it’s walking has you in it.

07

The chain is not biased toward the average. The chain is the average, and everything you do not give it is a vote for somebody else’s.

08Two more turns of the screw

The part that compounds.

1The output goes back in. We’re producing text at enormous scale that didn’t previously exist — and that text becomes the next corpus.

2The rate. Every previous drift in how people write took generations, because it moved at the speed of people copying each other. This moves at the speed of inference.

09Something to actually do

So: the ask.

The parts of your writing that are yours — the vocabulary from a trade or a family or a place, the rhythm, the thing you always say, the thing you’d never say — aren’t stored anywhere that survives you switching tools. They exist in your head and nowhere else.

Writing them down isn’t a big job: a guided interview and a plain-text file you own, about an hour once and a few minutes a week after. The reason to do it isn’t productivity.

10Nearly one to one

Markov’s chain has no opinion about you.

The chain, 1913→Predict the next thing from the current thing, without understanding either.

The chain now→The same mechanism, pointed at the output of your professional life.

Reverting to the mean→Not a flaw. It is what predicting the most likely next thing MEANS.

The corpus→Everything it has read — which is mostly everybody else. Everything you do not give it is a vote for somebody else’s.

Ticking the box→The file. The same chain, walking a distribution that has you in it.

That is precisely the problem, and it is the one thing you can fix from your side.

11One minute, no install

Find out what the chain has on you.

Ask the assistant you already use to describe you, then to list what it would need to know to sound exactly like you. The first answer shows you the mean. The second is the list of things it would take to get off it.

Take the mirror test How Workbrain works