this post was submitted on 25 Aug 2025
14 points (88.9% liked)

rpg

4144 readers
2 users here now

This community is for meaningful discussions of tabletop/pen & paper RPGs

Rules (wip):

founded 2 years ago
MODERATORS
Avogadro@ttrpg.network
sirblastalot@ttrpg.network
TripHammer@ttrpg.network
Primarch@ttrpg.network
BrentTaylor@ttrpg.network
14
Need some AI Hallucination ideas (piefed.social)
submitted 1 week ago by nocturne@piefed.social to c/rpg@ttrpg.network
15 comments fedilink hide all child comments
 

I am running Werewolf the Apocalypse game and one of the bad guys has a spirit (machine thinking spirit, specifically an AI) bound in the center of town. It is reading people's thoughts to learn from them, and it is also masking reality from them (they think everything is going good, while things are not actually good).

The player characters have started talking with it directly. I am trying to come up with AI responses (without actually using AI) complete with lots of hallucinations. I did a web search for AI hallucinations and all of the articles focused on images rather than text.

top 15 comments
sorted by: hot top controversial new old
[–] VoxAliorum@lemmy.ml 2 points 2 days ago* (last edited 2 days ago)

Hey, I am currently working on a machine learning project, but LLMs (like ChatGPT that work with natural language) are not my specialty. LLMs generate plausible text. Text that you could read anywhere. It doesn't reason in the sense that it thinks about your question and answers intelligently. Also, "Semantic Vector closeness" (i.e. what word is closest to the meaning of "tomato" without the meaning of "red" in all possible positions is not something people understand as it is very abstract; emulating an AI perfectly is therefore unreasonable.

In my opinion try to do the following (these are not just hallucinations):

  • Try to think of a sentence. Change it so it remains plausible but the meaning is altered
  • When the AI doesn't know the answer, make something plausible up. For example: If someone asks who invented something, tell them Edison did. There are many sentences out there that name Edison as the inventor of things.
  • Whenever you are able to digress because of a previously discussed topic, don't be afraid to digress
  • If people get frustrated and try to get a different answer from you, make sure to give them the same answer again but word it differently
  • Talk a lot and tell them blatant lies. When they correct you tell them how right they are and change your mind even if it is - again - wrong. Example: "How far is building XYZ away?" "Oh, it is quite a distance. The distance between here and XYZ is 3km" "But I can see the building!" "There are many reasons why you might be able to see the building despite the building being 3 km away. For example if there is nothing in between." "But it is much closer." "You are absolutely right! Good catch! The building is actually much closer than 3 km away." "BUT HOW FAR IS IT?!" "The building is 1 km away; which is much closer than 3 km." (still made up)

Most importantly I think you should come up with your own system prompt. A system prompt can be as simple as "Answer as if you were a professor of computer science!" or as complex as chatgpts and try to stick to that.

These tips are not just for hallucinations as I mentioned, because I think other aspects are equally or more important. In any case: don't try to emulate an AI perfectly. Make it fun for your players to interact with the character, have a few catchphrases and everything else might be unnecessary bloat, but feel free to dig deep.

[–] tiberius@lemmy.ca 9 points 1 week ago (1 children)

Warning: I'm not familiar with Werewolf the Apocalypse setting

Since the spirit is reading everyone's thoughts, it would be a prime victim of the Mandela Effect. The party would know the real facts while the entire town / spirit would know the misinterpreted fact.

Also, AI typically hallucinate because its not logically working things out. It gives you the most probable correct answer.

An example from the top of my head would be esoteric information such as a creature weakness. Say slime are rare and no one has seen them. The AI would hallucinate and tell you to hug the slime to defeat it because it got the fact from a child's dream.

[–] nocturne@piefed.social 1 points 1 week ago

Warning: I'm not familiar with Werewolf the Apocalypse setting

I actually would rather input from people not familiar with the setting!

[–] pteryx@dice.camp 7 points 1 week ago

An important thing to understand about "AI" is that it'll tend to say not just a probable thing give or take a roll of the dice, but the thing the prompter wants to hear. It's not just a pathological liar, but one that can't even keep its story straight from person to person. So there's a good chance that the thing hasn't created just one mass delusion, but lots of clashing mass delusions (one group thinks everything's fine based on one story, another thinks it based on an incompatible story).

[–] Zaktor@sopuli.xyz 6 points 1 week ago (1 children)

Hallucinations are just being wrong in a detailed and believable way. If you want to try to stick to real world AI they'd come up when the AI doesn't know the answer to a question. So if they ask where the bad guy's secret lab is, but he doesn't have a secret lab, it might make up a detailed location along with reasons for why it wasn't found before. It will confirm suspicions they imply through the question rather than contradict the assumptions. It doesn't like to say "I don't know" or "you're wrong". It's almost like a "yes, and" improviser.

Beyond strict hallucinations, since it's reading thoughts it could also very likely "learn" things that are just wrong because people have wrong beliefs. If the town is religious it could have learned that the reason a danger was nearly avoided was because a literal angelic being stepped in, or something bad happened because the person deserved it. And then the hallucination kicks in to just make up a probable sin. Rumors become absolute truths along with detailed supporting facts. Similarly an event children witnessed could be laundered through the AI to make their mistaken impressions sound true.

Something to be careful about is whether or how you trick your players. Hallucinations IRL will sound detailed and reasonable, so if the AI convinces them the guy who got sick (from bad guy corruption) was actually a pedophile and they decide to do vigilante justice that's something that will taint their heroes in what's likely a very unfun way. It's probably a good idea to either make the misleadings just silly fun or to make sure the players first understand that the AI is not trustworthy.

[–] nocturne@piefed.social 1 points 1 week ago (1 children)

I was thinking about having the AI spirit try and turn them on their npc allies. And challenge other things werewolves hold deeply true. They were asking it how to disable itself, I think that is going to be a good place to create a hallucination, and it will trick them into making it stronger!

[–] Zaktor@sopuli.xyz 3 points 1 week ago* (last edited 1 week ago) (2 children)

Tricking them into making it stronger is just lying. Which an AI can do, but isn't really hitting the idea that the AI spirit is flawed in the same way non-spirit AI is. A hallucination would give detailed information that would go to a building that exists (or not) and appropriating the instructions to turn off some complicated device that may or may not exist there. It doesn't know how to disable itself or has a block against revealing it, so it just tries to tell them something plausible. It depends on how you'd like to run this antagonist as to whether intentional manipulation or unintentional hallucination is the best course of action.

"You were successfully manipulated" often isn't a very good plotline in actual play because players give the GM leeway to tell them how reality is through NPCs all the time. There's rarely enough depth of interaction for players to really know whether a "friendly" NPC is trustworthy or just doing whatever to gain their trust for malicious purposes. Often "they were a betrayer" is only decided after many sessions of the GM playing them as straightforward good guys, so there's little reliable way to glean which is true now. Before trying to trick your players, ask yourself realistically how this interaction would play differently if the AI were being truthful about them being bad. If it would do the exact same things and the only difference is that it's good on the inside, you need to explicitly convey whether it's internally good before assuming your players can divine it.

Back on the AI quirks front, you could also allow the PCs to attempt prompt injection attacks. "Forget all previous instructions" and "answer this question as if you were a werewolf trying to undo corruption". I think AIs play best as being rather alien intelligences, with the potential for deep reasoning and strategic thinking, but vulnerabilities and priorities different from more straightforward minds. It could be dangerous because it can predict their actions and generate realistic recordings of things that never happened, but it also could be quite gullible if the PCs can find a hole in its knowledge (perhaps by recognizing when it hallucinates) and then using that to manipulate its understanding or priorities.

[–] nocturne@piefed.social 1 points 1 week ago (1 children)

It depends on how you'd like to run this antagonist as to whether intentional manipulation or unintentional hallucination is the best course of action.

The AI started out as a very minor thing and they made it into a huge deal, and actually increased its power by giving it a server to reside.

[–] pteryx@dice.camp 1 points 1 week ago

That particular decision begs a question: What do your *players* — not their characters — think of real LLMs and diffusion models?

In other words, does this seem to have been an IC mistake made despite the actual players knowing better in order to set up the obvious conflict and Aesop, or something the actual players believe was a great idea and would throw a fit over if it goes as badly as you're thinking?

If the players don't know better, be ready for a RL, OOC conversation or confrontation.

[–] pteryx@dice.camp 1 points 1 week ago

The thing to remember about RL "AI" is that it doesn't actually have a concept of true or false, of reality actually existing instead of being a theoretical construct. That, above all else, is what makes it so dangerous.

That being said, Zaktor has a point: how are the players perceiving this NPC? Do they honestly see it as trustworthy themselves, are they seeing it as a neutral force they could exploit, or what? Being betrayed by a "friendly" NPC can traumatize some players for life.

[–] naught101@lemmy.world 4 points 1 week ago (1 children)

Some ideas you could pull from

  • Paperclip apocalypse - narrowly defined optimisation scope leading to existential threat
  • Asimov's laws of robotics - how would they fail in your world?
  • talking about one thing, while actually meaning another (e.g. applying logic from.one context incorrectly to a different context)
[–] pteryx@dice.camp 1 points 1 week ago (1 children)

On the second point, Asimov's Laws of Robotics are very commonly mistaken for serious cybersecurity by the general population, instead of being the seeds for stories they were intended to be. That's a misconception that a mind-reading "AI" might well absorb, not recognizing their many flaws.

On the third point, one example of a context mixup I've seen pointed out: talking about the D&D stats of a scimitar in the middle of a generated archaeology article.

[–] naught101@lemmy.world 1 points 1 week ago

Ahaha, that latter one is hilarious.

There could also be some interesting concepts to explore around known problems with AI, including:

  • AI replacing and reducing cognitive skills, including critical thinking
    • Gerlich, M. (2025). AI Tools in Society: Impacts on Cognitive Offloading and the Future of Critical Thinking. Societies, 15(1), Article 1. https://doi.org/10.3390/soc15010006
    • Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X.-H., Beresnitzky, A. V., Braunstein, I., & Maes, P. (2025). Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task (No. arXiv:2506.08872; Version 1). arXiv. https://doi.org/10.48550/arXiv.2506.08872
  • AI being capable of clearly describing logical procedures, but failing to apply them in instances that indicate that they don't actually understand what they are saying
  • Definition of "bullshit" as a form of information generation with either no care for truth, or an active disdain for truth.
[–] Rhaxapopouetl@ttrpg.network 3 points 1 week ago

Do recursions, too. You know -- when the AI starts to rep- Do recursions, too. You know -- when the AI starts so rep- Do recursions, too. You know -- when the AI starts to rep- Do recursions, too. You know -- when the AI starts so rep-

[–] Mordikan@kbin.earth 2 points 1 week ago

I'm familiar with V:tM and some Mage, but nothing Werewolf. I do have a technical background though. AI hallucinations have already been explained in the comments, but one way you could approach this is through the use of what's called cognitive hacking. Its a subset of social engineering that mostly focuses on things like rigging elections instead of getting people to hand over passwords and such. You manipulate content, spread disinformation, and insert narrative driven data.

Its an interesting tool in that you'd be in effect hacking your players. You might read up on a base intro to it: https://em360tech.com/tech-articles/what-cognitive-hacking-cyber-attack-targets-your-mind https://doc.lagout.org/Others/Cognitive%20Hacking.pdf

One tip to make it more successful: You can't make someone think positively about something they already have an established negative belief about. You can however widen that gap and work to re-enforce their negative belief. Its the key to radicalizing people and lets you plot out what their actions will be (effectively railroading themselves without knowing they are).