McSpazz

I am aware. That means that you have 32k tokens to try and establish the context and characters. Every qualifier, every detail, every single one takes up tokens. The more characters, the less it has to work with. Incidental NPC's still take up tokens and potentially memory or, alternatively, they don't and any important details they mentioned are completely forgotten and you have to literally remind the AI as to the reality of things. The more you establish ahead of time, the fewer tokens and the less flexibility it really has. None of this changes my criticisms of AI's propensity to forget details, hallucinate, and just generally use stilted dialogue and struggle with concepts that are well outside of its training data. If you train the model using lore from CoH, it will remember that lore very well. However, if, in the course of an RP, you somehow do something that interferes with that training data, it will begin to struggle to retain that the longer you go on. As I said, the only thing that matters is the training data and the prompt. Generative AI, as of now, cannot adaptively learn things as the information it has to balance is altered. Put simply, the current technology does not allow for clean, consistent dialogue. I'm not interested in arguing what generative AI can become because I'm trying to advise people on where AI is right now. Put simply, based on how things are currently looking, I would be astonished if generative AI is capable of becoming any NPC with a given set of training data and world info due to the root data used for training starting to run thin and barely able to support what is currently cutting edge. I mean. Okay. But you would still have the exact same issues with generative AI that exists single player. Again, I've tested this out and found that I can flat out say: "X is true", get it to confirm X is true, and then some time later it says that X is not true and never has been. You could run a freeform Discord text RP with several friends set in City of Heroes and use generative AI to spit out descriptions for areas already in game (if that happens to be part of your training data). Sure, you'd have to write the NPC's, but they'd probably be more interesting and sensical than what a generative AI could randomly spit out.

I mean, okay. But at that point, you aren't roleplaying in the collaborative sense. You aren't participating in roleplay in the sense that people in CoH do. What you're doing reminds me of people who play RPG's and write their own little head cannon of what's happening in the game in their head. That's fine and it is, in a sense, kind of roleplay. But I think the vast majority of people here who have issues with AI in roleplay aren't thinking about roleplaying with literally nobody else and just a chat bot. We're talking about the collaborative practice of roleplay. The kind that generally requires a person on the other end. The kind where the quality of your own writing is just as important to your own experience as it is everyone else. If you find an AI chat bot preferrable to the company of other people, I shudder to imagine what kind of drama you've been through to make a soulless bot preferential to the potential of conflict.

I think what Greycat was getting at is that the responses usually have some weird qualities that stand out the longer you use the bot. The responses are all based on the training data and patterns and the more organic you want a character to sound, the more processing power and memory it requires to work out. The result is that, more often than not, the options most people have available for AI text generation is going to, like Crystal said, sound canned.

I just want to reiterate something after Crys' post. My issue with AI is not, in of itself, its usage. Again, ethical concerns aside (for which there are many), I can see many uses in roleplay. Acting as a springboard for ideas, helping form a foundation for a concept, describing scenarios that are unimportant to the core scenario, and so on. My concern, if you want to boil it down to its absolute core, is that AI is insufficient as a long term solution to making up for a roleplayer's own deficiencies and can prevent a roleplayer from attempting to improve in where they need the most work. Just as an example, part of what makes hands so difficult to draw isn't just their multi-jointed construction but also how even small errors in establishing depth can entirely bork their presentation. Roleplay isn't just a creative medium, it is also one that doesn't allow for long pauses (at least where MMO RP is concerned). MMO roleplay, in terms of how quickly you can access and utilize your skillset, is unmatched. Tabletop game roleplay might be immediate, but there is generally more allowances given given the physical presence or simulated presence at play. MMO roleplay hits kind of a sweet spot of having time to consider how you are going to reply and also not having time to spare to write out your answer. That's why substituting AI in roleplay is so hazardous and why I am being so insistent on people learning as much as they can without assistance from AI. The skills you can learn from roleplaying without AI are invaluable for a creative writer and are often essential to a roleplayer's development. If you have no interest in improving as a writer in every respect you are capable, I struggle to understand what draws you to the hobby of MMO RP in the first place.

AI has existed for, what, two years now? Literally, not long ago, if you did not like doing something in art, you either did not do it or you tried to get better at it. That was it. If you sucked at drawing hands, you either got better at it or got creative and found ways around not drawing hands. If you hated giving NPC's backstories, you could limit the number of NPC's you created and focus on making a few really good ones. On and on. You improvised, you found ways to adapt to your limitations, you got better. All of this improved more than just your skill in one thing. Learning to do things you don't like doing can improve your ability to do things you do enjoy. And if you couldn't do that, you literally did what you could do. If you only like to paint trees, you paint trees. Hell, I myself fall under one of your examples. I struggle with making and balancing NPCs. I don't fault people for using AI to make up for their deficiencies, but I think something is missed if you do. Learning to overcome those weak points, be it by getting better, finding work arounds, or simply finding ways to avoid them all together, are all important aspects of improving as an artist. Which is what you are. Roleplayers are artists. Roleplaying is a form of creative writing and even improvisational acting, both of which are artistic disciplines. Which is why I find the argument that this is just a hobby and just for fun so weak. Because, yeah, it is, but who the hell does a hobby or does something for fun but not actually fully participate in it? Something I've seen brought up time and time again so far is that the problem is when AI takes the place of the creative process and your own input, but that's what I've been arguing from the start. If tracing images is a tool for artists, why do artists get angry when an artist is revealed to have traced everything? It's because they weren't really participating in their art. Because, after a certain point, just tracing images doesn't really teach you that much and you aren't developing your own style. The same applies to roleplay and simply using AI to bypass having to learn to do things you don't like means that you are actively doing things to bypass the entire learning process. My line for AI in roleplay is filling in things that aren't overly important or completing overly repetitive tasks. Is the janitor's closet an important room for you to describe from ceiling to floor? No, probably not. I'd be fine with AI being used for that. The office of the big bad guy? The details in that room will actually matter and that's when it's important that a roleplayer knows how to describe a room. If you don't know how to describe a room well, how can you know when an AI does it poorly? How do you handle hallucinations or misunderstandings? We aren't talking about a sewing machine replacing seamster. We're talking about people using a technology to try and replace the need to utilize their own creativity. Something you can only enhance if you actively express and challenge yourself. Utilizing AI for writing in RP is actively going to hinder your ability to get better at. I've seen the kind of stuff AI can generate when someone tries to use it to make up for their own shortcomings rather than improve on them. It's really obvious and, honestly, I'd prefer they tried and sucked than give me something I can tell they barely even edited if not barely read.

I agree with your comments related to AI being able to serve as a better Google in some instances but I take some issues here. If you want to get very literal, yes. Telling an AI to take two existing styles and blend them together does make a "new style". However, remember that AI is incapable of generating anything that it hasn't been given training data on. That's why I do not believe that this is, in anyway, similar to a human way's of thinking. Children come up with ideas from whole cloth all the time without any reference material and while people might say there's nothing original anymore, humans don't need to be trained to come up with new ideas. They just do it. The creativity of an AI model is entirely reliant on the data it is fed. I'm not even arguing that you can create beautiful images with AI. My original post was strictly talking about how AI works and what I believe to be its limitations as far as roleplay is concerned. I'm not here to argue if AI is art or if AI images are easy to create. In short, very strictly speaking, my issues with AI images for roleplay is that you are likely going to struggle to get an AI image to match your character design if they have an outlandish design and you are going to struggle to make any image of that character consistent between images; especially if they have an outlandish design.

@Latex: I largely agree with you. As I said, the issues I have with AI art generation are mainly related to ethics. To clarify: I don't think you should RELY on AI for coming up with ideas. As you said, the biggest issue is when you use it as a substitute rather than a jumping off point. Also, I encourage people to commission. Obviously, not everyone can afford it and I respect that. There are definitely utilities to AI and using it for translation is definitely way up there. As I said, if you are not good at something, getting AI to do it for you will not help you get any better at it. If you don't like to write long descriptive paragraphs explaining the environment, you CAN theoretically use AI, but that just kicks the underlying issue down the road. AI does serve as a shortcut to help bypass things you don't like to do or are not good at in roleplay, but that misses the point of my core complaints with using them in the first place. If you just need a quick description for something offhand, fine. Your reference to long office memos is a good example of where AI might be well suited. However, if they're important to the story, I feel my points still apply. And, again, relying on AI for the core of your writing basically ensures that those aspects of writing will remain blind spots. Artists usually hate drawing hands, but you can be sure that if they can just generate hands, they'll probably never get good at doing it themselves. For the life of me, I can not recall any point in more than a decade I have ever seen a character that had no backstory. I've seen characters with simple backstories and characters that had basically no backstory and built it up over time through organic roleplay, but no backstory? That'd be a new one to me. I don't think the alternative to not using AI to generate a backstory is no backstory, I think it's more likely going to be a simple one that you end up building up over time. If anything, I think that not entirely fleshing out your backstory if you don't have a clear idea of the finer details is a good thing because it means that, if you become inspired later, you can fill it in without retconning. You can ask an AI to write in the style of a specific author or the speech pattern of some nebulous person ONLY if the training model has been trained on data that matches the prompt. The AI does not know why something is styled a certain way or why certain patterns exist. It only knows that certain data associated with the prompt you have given it has certain patterns. The actual style that it expresses is solely based on its training data. It is not, as of now, capable of creating a new style as generative AI is generated off of its training data and its training data alone. In order for it to have any concept of style or an understanding of what flows well, it would need to have generalized intelligence which is not what's going on here. The exact same thing is going on with AI image generation. The AI generator has no idea what it means for something to be in the style of, say, Picasso. It doesn't understand his motivations for drawing things in certain ways and it can't generate any images in Picasso's style unless it has been fed a large number of images that are in Picasso's style and has a set of images that match the rest of the prompt you want the item to be created in using that style.

After some consideration and some recent experiences, I've decided that it's worth tackling this hot button issue: AI generated content. Not only is it not not going anywhere, but the places it can and is being utilized in RP communities is only going to increase. I'm not going to get into the ethical debate about AI generated content here. Instead, I'm going to focus on three things: a very broad description of how it works, why utilizing it for RP is often not practical, and why I think you should generally avoid using it for anything you want to seriously attach to a character with those difficulties set aside. How it Works Before I can break down the hurdles of using AI generation in roleplay, it's important for me to explain what is actually going on here because a lot of what I am going to be describing later on is going to be very reliant on understanding the core of what's going on behind the scenes. AI generation, be it text or image, is, in very simple terms, pattern recognition. While it might be weird to imagine it this way given it is producing something rather than just reporting back patterns, that's, in effect, what is going on. It's just that the patterns are being saved in extremely complex data sets that are processed in less than obvious ways. It has a huge dataset, uses your prompts to figure out what data to use, then uses that filtered data to figure out an output that matches it. Text is a bit easier to grasp, so I'll start with that. Certain words are more likely to follow other words based on the subject at hand. The different tones you get when writing an email to a parent compared to writing an email to your boss is an example of a thin slice of what I mean by this. Certain patterns emerge when you look at text broadly and assigning certain text a label that describes that context can allow AI generation to group together patterns that match your prompt. A prompt focused on a movie description is more likely to include descriptions of excitement, glee, awe, and spectacle while a prompt for a eulogy is probably going to avoid those things. This is the reason why it is possible to do the process in reverse to determine the likelihood if a given set of text was generated by AI. The more the given text lines up with the highest probable to appear next word, the more likely it is AI. Images and even video are far more complex. With these, you aren't just looking at text but color, shapes, even art styles. The AI isn't storing an image so much as it's storing kind of an impression of what that image is with each image being associated with words that can then be compared to the text prompt. If you provided a prompt that included the tag "in a comic book style", it would utilize training data that has been associated with comic books. Why It Kind'a Sucks for Images I'm actually going to start with image generation here because, in broad strokes, it's more useful for roleplayers than text generation is. As I made clear at the beginning, I am setting ethical concerns aside for the sake of keeping the comments on this post from becoming a COMPLETE mess. However, even with ethical concerns aside, most AI image generation has a very critical flaw that is especially a problem for the wild world of City of Heroes. It only knows what it is trained on. AI image generation has no issues generating images of characters that fit standard designs or have common features, but it can start to really struggle once you start moving away from the norm. In fact, the stuff that it outputs can often really show the biases in the training data. Just as an example, an SG I was in was using AI image generation to make ID pictures of our characters and while the person doing it had no problems making a guy look really scruffy, it had a really hard time generating a woman with light scarring. That means that its ability to generate an image doesn't just change from one image generator to the next, but also between art styles within the same image generator. The more you struggle to get an image close to what you want, the more likely you will be unable to accurately recreate the character's appearance in another pose and/or setting. That means that when presenting these images, not only does it become fairly obvious you used AI generation, but it can lead to a very imprecise idea of what your character actually looks like which is generally the point of having the reference image/art created in the first place. Because of all of that, images you produce are likely going to struggle to maintain a cohesive vision in terms of style, fine details, or precise character design. Even small details in an image can lend itself to a certain feel or style. You can tell an AI to make something in the style of Tim Burton, but everything it gives you is going to be, in some way, based on existing images containing Tim Burton's style. It can't diverge from what exists which means that trying to implement your own twist won't be possible. However, as I said, it's not terrible at making reference images for characters. It's just that you're really not going to be able to keep the design consistent from one image to the next. Why It Really Sucks for Text The bigger issue I have and caution against in most circumstances, however, is text generation. As I explained earlier, all AI content is generated via pattern recognition. Each word that is chosen is determined not by any concept of style or understanding of flow but, rather, what the algorithm has determined to be the most likely to come next. This can produce text that, often, on the surface, looks totally fine. However, AI text often has quirks that make the output kind of weird to read or even entirely miss the point of your prompt. I could go over all of the smaller particulars of the AI generated text for fiction, but there are two that stand out. The first is that, as I said before, AI only "knows" what it is trained on. If you prompt it to write something that is extremely uncommon to exist in the written form, it's going to struggle. Just as an example, infecting someone with a virus to kill another virus but remain in the body afterwards and not actually cure them is a REALLY weird scenario. The vast majority of content the AI would pull from to write that kind of scene would be in medical settings and the idea of treating someone with a virus without curing them can end up confusing the AI, resulting in contradictory sentences. The second is that AI, at the time of writing, doesn't really have a memory. If it doesn't exist in the prompt and its training data doesn't have a clear idea of what to put next (such as when it is creating an explanation for something from Wikipedia), anything it generates may or may not be referenced at a later time in a way that makes sense. The longer the output the more likely this is to happen. AI cannot comprehend Chekolv's gun unless it was told to utilize it in the prompt. The prompt and the training data are all that matter. That means that the generated text can repeat itself, bring up concepts and then never use them again, or utilize far more detail than it needs to because it only recognizes patterns and humans sure seem to like their adjectives. Not having a real memory also makes it impossible to do any kind of follow-up story that makes full use of the story that came before and is entirely incapable of making all characters retain a consistent personality unless you are very careful about how you explain them in the prompt which can become next to impossible if other people's characters are involved. Why I Recommend Against It Even if you were able to overcome those major hurdles I explained above, there is a far deeper reason why I feel you should avoid using any AI generation for your roleplay. Again, even setting ethics aside, I don't feel it's in your best interest to get into the habit. Put simply: you cannot learn how to do something well if someone else is doing something for you. I know not everyone is an artist and some people just really, really want reference pictures for their characters. I still encourage people to learn to draw themselves or commission someone who can, but, as I see it, the only major arguments against ever using AI images aside from the issues I mentioned above are based around ethical considerations. However, text generation is another matter entirely. We are roleplaying in a text based medium and, as a result, the core of what we are doing is writing. Writing your own stories and text is important for the same reason that tracing images is frowned upon in the art world. You aren't really understanding why certain things are done and, also importantly, you aren't actually formulating your own style. Quirks for characters, little details about their possessions, what their home looks like that sets it apart from others, all of this can get totally missed if you just throw together prompts and literally rely on the more likely scenarios to carry you through. Lastly, of course, is that the discovery you are generating text for roleplay could be considered immensely insulting to everyone else involved. If you are not clear from the get go that you are generating and (hopefully) editing the text, it could appear that you aren't interested in fully engaging with the other players or are just doing the absolute minimum. Everyone else has shown up and are putting in the effort to try and build something together, why not you? In effect: why roleplay with me at all if you aren't going to be writing anything? Conclusion Look, I'm not your mom, dad, uncle, or even your boss. In all, I don't feel the time saved by generating text with an AI is worth the trade offs and you'd be better off saving up for a talented artist to draw your character to your specifications. All of this, once again, before we even get to the ethical considerations. If you insist on using AI, try and use it in a limited scope. Maybe use it to spitball ideas or, if necessary, help establish a starting point. Regardless, you'd still be better off as a writer and roleplayer to just do it yourself. Your ideas are worth more than the most likely scenario. This post is tangentially part of my series talking about roleplaying! You can see the full list of posts Here!

There's no "no redraw" option...anymore. As I said, if your character existed before they made this change, it could be forcing this onto you. The question is if this keeps happening after you set it to "original".

Actually this might be kind of expected depending on when you first made that costume/power config. The devs changed the way weapon draw works universally so that there is no weapon draw if there is a target for that power within range. After doing this, they seem to have removed no redraw as a customization option. As a result, however, any costume that had no redraw selected automatically set it to "Original", thus forcing the change.

On my Assault Rifle/Regen Sentinel, their weapon ALWAYS resets to the default, sheathed Legacy Assault Rifle when I enter the costume editor. It doesn't matter what weapon was selected or if it was sheathed/unsheathed.

The Tall Hedge (End) SG base item is incorrectly rendering the color/texture on the leaves sticking out of the item. As you can see, they are pitch black. when the preview image is not.

Even if I'm not in full agreement on the OP's method, I do think that more means of encouraging open world play would be really nice. ESPECIALLY blue side given how many zones it has that just go untouched.

To Hero 1, Dark Watcher, and Proteon or anyone from the original dev team that wants to say something! In the final year or so of the game's original life, the story really seemed to escalate in terms of scale. The stakes were ramping up as was the potential power of characters (both in lore and mechanically). How visible was the sword of Damocles over your heads in terms of awareness of the game's closure? Did it play any part in your decision making for where the game's story was heading? If you could go back and expand on any group that you feel didn't have much content made about them, who, why, and what would you want to do? For anyone on the original team! What was your reaction after hearing that Homecoming had finally gotten officially licensed?

With a team as small as Homecoming's, if you are going to implement new features, it's usually best to do so in such a way that it either enables the creation of further features down the line or makes future development easier. That's why I pointed out that finding a way of imposing SKing down via the zone/area you are in could be beneficial to revisiting enemy groups. It's also why I think that implementing new ways of interacting with zone events as Errie described seems like the most likely as it is expanding on something that already exists, could enable new functionality, and requires fewer new systems than trying to balance outright zone wide exp sharing.