Don’t forget we moved!
https://brandmu.day/
Concordia Thread
-
@mietze This is just my experience, but when I’m using an AI program to help create a desc, be a room or person, or item, the result always tends to have this impression that it’s describing something and telling the reader how they feel about it. Or should feel.
Sometimes you get weird stuff. And you’re given a flood of verbs on how you, as a reader, interpret something.
-
An inescapable comparison to this game is Arx, where GMs have been run ragged trying to keep up with our need to be catered to. I can’t begrudge these people for experimenting with a strategy that might be more sustainable for them.
-
This is very interesting to me.
When I see stuff like that I tend to assume that the person is just perhaps abusing their thesaurus a bit.
I do tend to write a lot of detailed and descriptive scene sets, flavor poses, and descs so now I’m wondering if people think I’m lifting them off AI! Or maybe they don’t think that because I don’t metapose in a directive way in them.
-
@mietze I checked your description before I ran into my daily limit. You are PROBABLY a HUMAN.
MAYBE.
-
-
@helvetica I hope the roster has a good balance of older PCs then. What ultimately did me in on Arx post-Ari was that the roster skewed young.
-
@hellfrog said in Concordia Thread:
I think using ‘AI’ is going to be pretty obvious but also not really - a problem?
This should probably be (yet another) thread split off of this one. Without going too far off on a tangent, I would argue that the “problem” is in normalizing use of these tools, which objectively harm creatives.
-
@Faraday There’s always room for more AI discussions in life, lord knows, but MOST of the posts are particularly about the use of AI on this game. I’m not sure why you would suggest splitting it – and I’m pretty split-happy as a person.
-
@Tez I feel like I have an idea where @Faraday is coming from. (Faraday, apologies if I’m putting words into your mouth. Not intending to do so here.) I had a post in mind to reply to this thread, but my points related just to Concordia were super brief, and I realized what I had to say quickly began going diverging into general discussion about the uses of AI.
(Isn’t there already a thread about uses of AI on this forum? I’m a little zzz so I’m too lazy to look right now.)
With respect to Concordia specifically, I’m not really against the use of AI to “automate” creation of content such as roster characters’ quirks and descriptions in particular, but I would draw the line at using it to generate the content for explaining the game’s theme and its underlying details. (Unless the game literally was made to showcase the content being written by a generative AI, anyway.)
I do concur with Faraday about the impact on creatives, but that’s a subject for discussion in a general topic on generative AI, so I will stop there.
-
@dvoraen There are a couple threads:
https://brandmuday.mythicus.net/topic/316/would-you-roleplay-with-ai
https://brandmuday.mythicus.net/topic/355/ai-megathread/
I’m p interested to hear what people think about AI-generated text in games, so I totally hope folks take the more general discussion over to one.
-
Concordia’s staff did address the issue on-game just a few minutes ago, probably wisely choosing not to engage here:
Spes
Hello! While we tend not to interact with feedback that has not been brought directly to us, this is one thing I do want to address because it is important to members of our community, and because we never intended any deception on the matter.
In addition to Midjourney, which we have used for every bit of art on our game, I do also use ChatGPT as an editor for content that I write. AI did not come up with our concept, our themes, our setting, or our story. If it did, it might have less holes and inconsistencies! But something Aequitas and I would like you all to understand is that we expected maybe a handful of our friends to poke around this place when we opened. We never expected the overwhelming response we’ve gotten, and we are so excited to have you here and to tell stories with you.
I want to be clear that sometimes, my contributions to those stories may have been rambled out into a Google doc at 3am and then handed to ChatGPT to make coherent, correct typos, and unify language/tone for a cohesive experience. It also allows me to give consistent quality whether I’m having a high inspiration day or a fried-brain sort of day.
I understand if my use of this tool, even as an editor, creates discomfort or causes your interest in the game to wane. We knew from the start that this place would not be everyone’s cup of tea, and I want to thank you regardless for checking out our game, contributing to our awesome community, and providing us with invaluable feedback.
Moving forward, if there is something that doesn’t sit right with you, I want to reiterate that feedback and discussion offered in good faith is welcome and encouraged. But if you want us to see it, please bring it here directly, to the request queue if it’s private or the forums if you feel it warrants community attention and discussion.
Thanks so much for your consideration.
-
But. LLM don’t ‘edit’. They only generate. That’s all they can do - generate off prompts.
fr fr
(she/her) -
@hellfrog said in Concordia Thread:
But. LLM don’t ‘edit’. They only generate. That’s all they can do - generate off prompts.
This isn’t strictly true, as I understand it.
You can insert a text and ask it to edit it, change it, stylize it, without adding anything to the actual content. You can describe Spider-Man and then tell it to ‘make the necessary changes so that this uses the British spelling for words’ for example, stuff like that.
In Occam I trust.
-
@Coin But I don’t want a description of Spider-Man.
ANYWAY. The AI stuff is kinda whatever, but I’m glad they acknowledged what they are doing and we can move on. In the future, I think games should be open about having an AI policy because using it isn’t bad or anything, but being transparent is always a good thing.
-
@Coin I’ve asked it to 'edit" some text blocks and all it does is paraphrase or rephrase. Granted, I wasn’t asking it to change things to a british spelling.
Either way, I don’t think someone is bad or even wrong for using it to generate IC text for their game, I just don’t want to engage. So transparency is the ask.
fr fr
(she/her) -
@Coin said in Concordia Thread:
You can describe Spider-Man and then tell it to ‘make the necessary changes so that this uses the British spelling for words’ for example, stuff like that.
Sure, but it’s doing that by looking through its language model for “SpiderMan writing” and “British writing” and then generating something new from the mashup of those things and your prompt. (Oversimplified and imperfect analogy)
The best description for ChatGPT I’ve heard is that it’s a word calculator. All it does is figure out “what word comes next given the last input” in the same way that a calculator solves one numeric operation at a time. Can you string things together? (7*(9+2)+4)? Absolutely. But it’s still at its core doing its calculation one word at a time. And so is ChatGPT. It doesn’t “edit” like a human does.
@dvoraen said in Concordia Thread:
@Tez I feel like I have an idea where @Faraday is coming from. (Faraday, apologies if I’m putting words into your mouth. Not intending to do so here.) I had a post in mind to reply to this thread, but my points related just to Concordia were super brief, and I realized what I had to say quickly began going diverging into general discussion about the uses of AI.
Yes, exactly.
-
I am just mildly confused… Why use AI to edit when Word/GoogleDocs/Spellcheckers in general have the capabilities to do that? To me, it seems like using something like ChatGPI or whatever version of the AI to ‘fix’ ideas takes away the personal touch a person gives. Like you’re telling it to ‘Fix this’ and from what I’ve gotten from the AI spit out is not really mild edits but complete rewrites with some common elements to what you asked for. I don’t really care if a person uses AI for their stuff, I am just confused to why use something for what many, many tools are already about for that has a different purpose.
The most I use AI stuff is to help me think of names for stuff like towns, cities, ‘groups’, and so on. OR a write up for a thing to get inspiration. Same as I use generators for.
-
I suspect it depends on how loose one’s definition of “edit” is. You can throw a moodboard of written vibes at chatgpt and have it “edit” that into a cohesive whole. I think most of us would call that drafting. I don’t know what the input was, here.
It’s a bummer, but it won’t drive me away from the game by itself. I’m here to play with other humans. Folks using AI for their roleplay is a total dealbreaker, of course.
-
@Rinel said in Concordia Thread:
It’s a bummer, but it won’t drive me away from the game by itself. I’m here to play with other humans. Folks using AI for their roleplay is a total dealbreaker, of course.
Waaaaay back in the day (we’re talking late 90s) there was an AI experiment being run by a PhD student in Canada where he was training an AI to MUSH. After having “her” play a character on a couple, he started up his own Star Trek MUSH to have a more controlled environment, but never told any of the players.
I remember being shocked when I found out, because I had no idea (I wonder if I looked back now, knowing how ChatGPT and such works, if it’d be better able to do a Turing Test) and never would have suspected.
Probably not as shocked as the guy who was TSing “her” on the regular, though.
(I only dodged that bullet because apparently I wasn’t even interesting enough to get an AI to TS with me back then. Talk about a hit to the ol’ ego.)
-
You can definitely improve your work by providing GPT with the text you have written, which will serve as the input for the editing process. (Requisite ‘I had ChatGPT edit my post for me’ joke. In this case, I let it fix absurdly bad grammar. Input: “Edit this into better language please: you can for sure get GPT to make some good edit on your work if you give it them things you did wrote first as what goes inside!”)
While the ChatGPT interface is designed for chatbot-style exchange, the underlying model (GPT-3/4) still involves feeding inputs through functions with parameters to generate a given output. You can give it anything for a prompt and ask for any length of output (or let it decide how much to give you), and you can control, in numerical terms, how much the input and output are correlated. If you set it to discard all but the most likely probabilities, you can get something that will resemble your initial text very closely (or exactly). Thus, while it isn’t going word by word or sentence by sentence and fixing things, it is instead creating a ‘novel’ text that may be 99% similar to the original.
This is easier to demonstrate with image generating AI, as they can operate with a text prompt or with another image. So you can ask for a dog, or you can give it a dog and ask it for a 95% similar dog. This is useful in photo editing as a blend pass on what would have previously been ‘obvious shops.’ But the idea is pretty much the same.
