So I recently had a puzzle idea that needed some bulk filler text. Not wanting to spend time writing text that is essentially fluff, I decided to try out some AI generated text.
Sites like https://app.inferkit.com/demo or https://deepai.org/machine-learning-model/text-generator take in a seed of text, then use predictive AI algorithms to continue the text. So how does this make for a useful game puzzle? Won’t it just generate a bunch of modern writing garbage unrelated to a fantasy world?
Here are the steps I used to create this collection of documents detailing the fluff of an organization the players were sent to investigate. The background is that these 8 pages represent the collected folder of documents handed to the group by their quest giver, abstracting things a touch because I don’t want to create that many real pages nor would the players actually read them.
Step 1 was to come up with a handful of document types that I wanted to create. I decided I generally wanted this to focus on financial documents, taxes, organization bylaws, and the like. So, with each document type, I entered in the starting sentence describing something like “The Dawn’s Charge financial obligations” or “The following are the bylaws of the Dawn’s Charge”.
Step 2 was to repeat that a few times for each prompt, gathering the generated texts that seemed useful. I spliced the gathered paragraphs together into a single page for each document type.
Step 3 was editing each of those documents to make them fit a fantasy setting. This means changing modern references to the IRS or United States Government to their fantasy equivalents: the Kingdom of Alara and the Hand of State. Make sure to give things a once over, you don’t want to leave in references to dollars, modern technology, or other similar things that would ruin the immersion.
Step 4 was to create the puzzle itself. In this case, I put page numbers at the bottom and top of each document. I then drafted a “cover sheet” in the form of a press release that hints at the solution to the puzzle. In this case, the solution is to use those page numbers to get the first word found on those pages within the Player’s Handbook.
The top numbers: Check (PG 55) this (PG 6) chapter (PG 45) while (PG 218) getting (PG 146) the (PG 47) thieves (PG 43).
The bottom numbers: The (PG 47) cleric (PG 56) careful (PG 102) acting (PG 86) for (PG 175) its (PG 260) works (PG 53).
Using the numbers hidden in the bylaws document: Fey PG109 Creatures PG170 Dungeons PG318 Project PG270
Step 5 was to add a few bits of more relevant fluff to the documents to tie it into the plot. This included name-dropping relevant NPC’s, tuning the press releases to mention current events, and tweaking some wording to set the scene of what type of organization Dawn’s Charge is. These wouldn’t be necessary to progress, but might be fun easter eggs for the players to find within. I wouldn’t suggest doing much more than that, otherwise they can easily become red herrings that throw your carefully planned puzzle out the window.
AI text generation is going to become a more and more useful tool in any DM’s toolbelt, so I’d suggest giving it a try next time you want to spice up your game with some props or puzzles. Don’t be afraid to generate two or three times as much as you want, then delete anything that doesn’t fit. The really amazing text generation results come from aggressive trimming, editing, and selecting from some of the real garbage that text generation can produce.
Have you used AI generated text or images from the GPT-2 or 3 neural nets in your games?
Descending into the dragon's cave The dark and cold gloom To wake up and take up what we've left I've come back out of them Till the moment their body was inside My heart beat, I'm not dead But I see myself Faced by a monster in that mirror Sewen with that curse I see myself as a monster in that night I come back out of them because now we've got a dragon inside I see myself and I'm not dead But I've come back out of them because now we've got a monster inside