Statement on AI generation, player descriptions and beyond

Most of you by now have already seen the results of this so-called generative AI productivity "boom". As I work in tech I've already seen MANY coders attempting to do way too much, way too fast, because they think AI is going to code everything for them. This echoes in creative works, with the word of the year being literally SLOP. Referring of course to the metric shit ton of lazily generated content that is being pumped into books, movies, video games, in pursuit of profit or often just to save time (and effort).

This post is not to try to discourage or encourage anybody to embrace or avoid AI, I'm simply telling you what AR is doing in this moment.


AR is a handmade game that is over 20 years old. It has been balanced by sweat and frustration of real people, by trial and error after error. So AR will remain an entirely hand-crafted world. We are choosing to keep AI out of AR's development. That does not mean that our imms (including me) would never use an LLM as a productivity assistant, but we have something 20 years old that has aged like a vintage fine wine here, warts and all, slop-free. No AI-generated code translated 3 times. No AI copilot "fixing" our game's bugs without a dev that understands. No AI written consortium stories. The only slop you find this game is in the ofcol pig stys.

----

This brings me to character descriptions which are entirely player generated. Of course, everybody recognizes that writing a description is an effort, and an LLM provides an easy shortcut, which as soon as someone else takes, its easy to follow them off that cliff. Recognize though that investment into a character is the way to foster a deeper connection to it, to make the world feel more immersive and real. The AI's can generate descs, backgrounds, lair descs, but it will not feel like its truly yours, and others who interact will no doubt notice it is soulless. Similarly, AI's can also review descs, which gets us into the position of AI doing double duty, a plain waste of electrical power, and nobody caring about the descs anymore. That being said, AR's imms do not have the resources to just ban AI stuff, and I know some of you don't really care about the descs too much.

We are using AI to push back on AI, with two key benefits immediately in effect.

1. As of now, if you visit your website player page, you may be able to see a list of "violations". These you can certainly take with a pinch of salt, as they are entirely AI generated. We setup an LLM with instructions to check if a description follows guidelines and provide suggestions for improvements if not. An Imm can kick this off for you on request, and they already have the tools built to do so, just ask in game if you need a run.
Note that only YOU can see the violation list given on the website.

To clarify this further, here is a very abstract example. Lets say you have a char whose name begins with a B. the link to the player page would be like this, https://abandonedrealms.com/players/display.php?name=b (finish the name). if the bot has checked, it will either show the issues with the desc underneath the desc or it will say, there isnt anything to see here. Since it is your character, only YOU see the issues. No other player will see that.

2. Our "gilded" descriptions process is now gated by an AI slop checker. Basically, the same bot we wrote to check your description will also take a guess if an LLM made it. If that is determined, we will likely not be giving a gilded description. The decision is a human one taken under guidance, so there will be exceptions where the detection gets it wrong and we just go ahead and give it the gold trim.

TLDR: we are not planning to ban or prevent AI descriptions - we have added anti-slop and provided in-house use of AI to self-check your description against our hand-written guidelines.

awww not my one stop slop shop!

paid someone on fiverr 20 bucks to finish Szrevan's final background story because i could not get in the vibe, that's a whole month of ChatGPT Pro!

stop yawning my posts fucking liberals

or imma fuck you up with a trash necro

m1co if I was wanting to stop players from using AI to write desc then why would I provide a tool that helps players improve their desc using AI? The goal is just you have to EARN gilded. That's HARD mode. Which is how it should be.

Mico, if you don't want your posts yawed, don't post stupid content. It's that simple.

I fail to see how one's political views have any relevance whatsoever regarding this. Which is why I yawned your two posts after that, because you're just overreacting without using your head.

PS. I fully approve of the measures being taken.

Well shit, when I try to focus on just physical descriptions I sound like AI.

* The writing style is clear and grammatically correct, but lacks unique and evocative phrasing that would typically indicate a human author with more effort. * The description uses fairly standard fantasy tropes without much individual flair. * The language has a consistent tone and structure indicative of AI text generation.

Maybe there is a complement in there.

The "better description" I had I worked on a while, was longer and bounced off off ChatGPT for feedback. Short descriptions where I focus on physical appearances without any help, I get looks like AI feedback, so complement or stop writing like a robot?

Yeah the struggles are there for me as well Grayden. I feel the so-called "improvements" that are suggested often make my character sound less human-produced based on feedback I've seen on my active characters.

I've noticed it doesn't like the active phase of action phrases, e.g. out of stare, stares, stared, and staring, it especially dislikes staring.

One character with a relatively short description I wrote had no feedback whatsoever, positive or negative so I guess it passed the test? It wasn't gilded and honestly I felt it was a more "basic" description. Sometimes my choices of words are seen as archaic as well (nothing wrong with that if you ask me).

I had a character who I modeled after a celebrity at certain stages in her life and the AI checker didn't like it, saying it was too generic.

OK just remember the only goal in mind for evaluation with AI is to pass the "AI slop" check (i.e. "was this AI generated"?), if you like the suggestions and they get you thinking, great, but they're not part of evaluation, and at the end of the day, you're still being peer-reviewed by imms who properly understand what passes for a good desc on AR. The machine does not have the nuance to know that a level 5 nobody character is not worth an 80 line desc, and it actually only flags for generic because the guidelines in "help description" warn against generic.

AR is a handmade game that is over 20 years old. It has been balanced by sweat and frustration of real people, by trial and error after error. So AR will remain an entirely hand-crafted world. We are choosing to keep AI out of AR's development. That does not mean that our imms (including me) would never use an LLM as a productivity assistant, but we have something 20 years old that has aged like a vintage fine wine here, warts and all, slop-free. No AI-generated code translated 3 times. No AI copilot "fixing" our game's bugs without a dev that understands. No AI written consortium stories. The only slop you find this game is in the ofcol pig stys.

I love that so much and am excited to note AR's stance on this in the next advertisement. AR is def something special.

I'll fess up and say I've been guilty of this. Agree with Kalist here, the game is aiming more towards improving human interaction, it's less human interaction if it's less of humans doing it.

Sidenote the disparity between the quality and usefulness for AIs is amazing. The one we've been allowed to use at work keeps refusing to even let me test it for certain things and just says to consult my legal office, even though we're being pushed to use it as a tool. Completely baffling. Or I'll ask it policy questions and it doesn't have access to those manuals and it just guesses based off open source information like Wikipedia

Well please dont get it twisted. We're not forbidding to AI generate your description at all, as long as they meet the rules and dont immersion break people with planting thoughts in their heads and the like. We're being more careful with handed out *gilded* - if it looks AI-generated to our imms, it just isnt worthy.

* The writing style exhibits some formality and a tendency towards very precise, almost clinical descriptions, which can be characteristic of AI. However, the incompleteness of the description and the slight awkwardness in phrasing also suggest a possibility of human authorship, perhaps a novice writer. Without more samples or context, it's difficult to definitively ascertain AI involvement.

Ouch, your AI called me a novice writer.

Wrote 2 whole ass descs from scratch and get accused of being AI lmao. Might as well just generate one with AI and call it a day.

youre reading an AI report, final decision to elevate is by a human whether it reports AI there or not in that page i gave to check. and dont worry the imms can also see it kinda sucks

Got it. As long as the bot doesn't automatically disqualify a desc for consideration, I'm good.

Damn. M1co is still kicking?