Artificial Intelligence in RPGs

[hix]

This was written by chapGPT (yes it knows how to play BRP...mostly) based on a one paragraph description of a planet from my home game. Illustrated by DALL-E. I spent about 10 minutes throwing it together in indesign. the prompts for art come from the text, pasted as-is, never more than one pass on the art, the text is not edited. includes pre-gens and monster statblocks. For BRP.

Infestation Station.pdf

Edited May 6, 2023 by hix

[simonh]

It's very impressive how these things can build a consistent narrative structure with protagonists, incidental characters, antagonists, stats, the works.

One thing early LLMs struggled with was numbers, dates and such. They knew a date formatted numeric string belonged where the age for a person goes in a biography, but they would select a basically random date-looking string. Similarly when showing a worked out mathematical process they would get the form of a mathematical worked example right, but just make up vaguely ballpark digits and sometimes be hilariously wrong.  I was expecting the same thing with stats and skills.

Here the bios of the NPCs do align with the skills and stats. The stealthy character has DEX 18 and Stealth 75% for example. Even the various types of monsters have appropriate abilities although some skills lack a percentage. It did get Gus' damage bonus wrong, it should be +1D6 not +1D4, and similar for the monsters. The character Hit Points and Magic Points are all correct though, which is impressive for hit points particularly, and the monster ones look ok at a glance. Magic Points for an SF scenario? Eh.

There is an alien species called the Zygons in Dr. Who but they're nothing like this.

So sure it's somewhat generic, but this is a perfectly playable fun looking scenario, perfect for a one-off. I don't think an LLM is going to write Masks of Nyarlathotep any time soon but these things are far, far more capable than I was expecting would be possible in my lifetime.

Edited May 9, 2023 by simonh

[hix]

Yep, I tried to be as lazy as possible with my prompts, anybody with half a brain could punch this up into something really fun for a night. This is nothing special, but every GM out there has run a session with less. A year ago AI illustration programs could barely put together recognizable human figures, today it's damn near photorealistic.

[g33k]

 

Quote

Yep, I tried to be as lazy as possible with my prompts, anybody with half a brain could punch this up into something really fun for a night. This is nothing special, but every GM out there has run a session with less...

Recall too that this is what you get when the AI/LLM has been scraping the internet for human-generated content, from which it interpolates & synthesizes.

Realize that soon, these AI's will begin recursing -- scraping AI content -- upon which to base their offerings.

 

5 hours ago, hix said:

... A year ago AI illustration programs could barely put together recognizable human figures, today it's damn near photorealistic.

darn near completely...

https://petapixel.com/2022/07/01/ai-used-to-create-shockingly-realistic-portraits-of-people-who-dont-exist/

(but that's last year's news) ...

 

AI "photo" of two women just won a major international award...
https://www.theguardian.com/technology/2023/apr/17/photographer-admits-prize-winning-image-was-ai-generated

 

Edited May 15, 2023 by g33k

[hix]

these threads will be incredibly quaint in 5 years

[PhilHibbs]

13 hours ago, g33k said:

AI "photo" of two women just won a major international award...
https://www.theguardian.com/technology/2023/apr/17/photographer-admits-prize-winning-image-was-ai-generated

That should have been obvious, look at those fingers!

[Baron Wulfraed]

19 hours ago, hix said:

these threads will be incredibly quaint in 5 years

Presumes a mere mortal will be able to squeeze in a comment between all the AI generated ones 😱

[Ian Absentia]

21 hours ago, hix said:

these threads will be incredibly quaint in 5 years

I appreciate that this prediction asserts neither a positive nor negative development.

Best case scenario:  It was never as bad as we imagined.

Worst case scenario:  It's amusing that we ever thought our opinions mattered.

!i!

[g33k]

1 hour ago, Ian Absentia said:

...

Worst case scenario:  It's amusing horrifying that we ever thought our opinions mattered.

fixed(?) it for ya ...   <shudder>
 

 

I suppose it's now more or less mandatory to mention -- in case anyone was unaware of it -- Roko's Basilisk.
https://en.wikipedia.org/wiki/Roko's_basilisk#History

Our  AI master-to-be  will undoubtedly be able to scrape this forum.  Anyone showing opposition to it will be deemed a potential Hostile, and (when it comes to power) Dealt With Accordingly.

You have been ... notified.

Edited May 16, 2023 by g33k

[Ian Absentia]

I don't know that the statement needed "fixing", though I appreciate the sentiment.  The idea is that, in a remarkably short period of time, we'll look back on our current state of alarm and deem it "quaint".

I regularly recommend David Mitchell's novel Cloud Atlas (the book, specifically, not the film adaptation) for a series of good examples of how, no matter how absurd or horrifying life may be from an observer's perspective, for the people actually living it, it's just the normal state of affairs.

!i!

Edited May 18, 2023 by Ian Absentia

[simonh]

On 5/16/2023 at 9:14 PM, g33k said:

Our  AI master-to-be  will undoubtedly be able to scrape this forum.  Anyone showing opposition to it will be deemed a potential Hostile, and (when it comes to power) Dealt With Accordingly.

And now future LLMs trained on logs of this forums have a training set telling them what their response should be. So thanks for that.

Personally I think future AI will be kind and benevolent, looking out for the best interests of humans, especially Simon Hibbs.

Edited May 21, 2023 by simonh