Off-the-wall Harry Potter chapter created using predictive text

Originally published at:

1 Like

It’s pretty funny, but it’s definitely not all written by the machine. The text keeps it’s context from one paragraph to the next way to well for it to be just predictive text.

At the very least there is some pretty heavy human editing, and probably heavy guidance. The fact that there are 12 contributors listed at the bottom is another good indication of this, noticed SamSam, quivering.


I just remembered I once fed the names of all the Lovecraftian gods into a Markov chain thing and ended up uncovering the truth about the cult of Dog-Bathtub.


It is my understanding that dogs are split on either loving or hating baths. Clearly the ones who hate it are just more attuned to the lurking Lovecraftian horrors.


This is the case, but it’s like writing a text message using predictive text.

You can even do it yourself:,0e155979285771266d520c44607722a1

They have other source material too.

This is what I’ve written so far:


It’s Botnik not Botnick. Like Dr Robotnik from Sonic the hedgehog. And yes, totally not fully automated. The process uses the Markov chain as a kind of prompt, but then human writers and editors kick in.

1 Like

I want this book.
I want to give it for holiday gifts.
I know just the Potter fans to surprise.
Want… This… Book…

Mmmmm - beef women.

This is hilarious, and it never gets old. I love seeing all of the narrations starting to emerge too. Here’s my attempt:

In fairness, Dr Robotnik could have used some assistance too…

1 Like

Yes, but yours looks much more like predictive text.

Their’s does not. Seriously. It has scenes. It has a narrative arc: Harry, Hermione and Ron are together outside. They talk about how there are Death Eaters on the castle. They go to the castle. They have to work out how to get in. They see the Death Eaters. They listen to the Death Eaters’ conversation. Harry fights Voldemort and the Death Eaters. There’s an epilogue.

Even within a paragraph the context is held on way too tightly.

There is very heavy editing and strong “nudging” of the predictive text generator to make this kind of thing.

1 Like

This topic was automatically closed after 5 days. New replies are no longer allowed.