There is a trope in films that I have always hated. Someone explains something complex, and another character cuts in with "please, speak English." It undermines what was being said, and suggests you should only say simple things to be taken seriously. The more I talk to Fable, the more sympathy I have for that character.
Fable writes in a way that has drifted away from how we write, and the skill that helps you read it is older than computer science.
How Fable writes
Leave aside the dashes and the tidy contrasts everyone spots; the difference is subtler. Fable barely writes like a conversation, and it almost never explains itself. Every word is "load-bearing": it reaches for the most information-dense word it can find, so there is almost no breathing room left in a sentence. And it writes as if it is talking to you alone, assuming you remember every word of the conversation.
Show that output to someone else and they tell you they recognise the words and have no idea what it means. I often have to edit Fable's output to make it legible, which is really a translation, and that is fine. It does raise a question about how we work with models that keep getting smarter.
These models learned from human language and conversation, and the smarter they get, the less they sound like us. Nobody trained Fable to write like this; it emerged, and emergence is hard to put back in the box. You can see it in the em dash. Write in capitals in a skill or in your custom instructions that the model must never use one, and it often still leaks through.
Half the meaning is elsewhere
The em dash is the harmless version. A model developing its own vernacular has a real effect on how we work with it, and that vernacular is shaped by the conversation you have had with it. That is why AI fluency comes from interacting with these models a lot, far more than from technical skill.
Fable writes densely because it assumes you were there for all of it, so half the meaning lives in your head rather than in the text. Hand the output to someone who wasn't part of that conversation and all they see is words.
An old way of reading
Reading dense, layered text is not new. Oracles, priests and imams have done it for centuries, and the discipline is called hermeneutics, which you won't learn in a computer science class. You pick it up from anything you have to read several times before you get it: a book like Dune, a hadith, a Jotie T'Hooft poem.
So I think the people who feared these models would flatten language and culture have it backwards. The more you work with them, the more it pays to read widely and get comfortable with difficulty.
It's a shame we have lost the em dash, but if it pushes us to stretch what human language and culture can do, I'm all for it, starting with a trip to the library.
