Jump to content

Danko Sedin

Vintarian
  • Posts

    22
  • Joined

  • Last visited

  • Days Won

    1

Everything posted by Danko Sedin

  1. Interesting video related to discussions about how steam actually handles AI slop. It seems, that for VS case - at least something need to be done with donations.
  2. I don't have a strong data about it, but my small experience with my only mod tells me that in order to grab a lot of downloads you need to do basically two things: first you need to "sell" your mod to server owners. Convince one guy - you get 30 downloads. and second - you need to update often. every update - you farm you servers for downloads. as long as wave become shallow - its time to release new update and farm a little bit more.
  3. I am sorry for that feeling. I see your position as an ethic, so I accept that it is something I can't change and wouldn't try. But I want to understand a background of it, and you gave an excellent description. I am grateful to you for that. And I was trying to give my background of my ethical position as well, I am sorry if it was like adding gas to the fire.
  4. Ok, now you can feel what it like to be assumed of something. I am sorry that I offended your feelings. The point I was making though is - no matter how picky you can be there always would be someone who could be more pickier than you. You mastered 3 areas? What about marketing?! Do you use CI to build your mod? What about writing code in only true style - functional?! It is never ever ending journey to learn something. Even for people who use AI. They learning AI, for example. Or game design, whatever. I can't share you feeling about code as art for one simple reason. I never seen somebody who wrote a line of code, stand back and said "The best line of code in my life". For me, anything written is a piece of sheet as soon as it written. Just because I always can do better. I think real reason why people feel that AI steals something from them(and being very protective of their code) - because AI made implementation cheap, but ideas valuable and in the same time very fragile. Before AI idea would cost almost nothing - you need to create it before it will obtain value. And code was a proof of creation. Now, when you can create whatever you want you need to want something cool at first place. Idea become something valuable, because you can implement it relatively cheap. And also other people can stole your idea and create something of their own. In almost all countries law will agree that their creation is their, and you wouldn't get anything from your idea. This is sad reality. As for the image - equation is enough. Color for this image usually defined by numbers of hopes that point need to go out of specific area. It is a beautiful Mandelbrot set if you want to know more about it.
  5. After tons of videos like "VS vs Minecraft" I've had an impression that Tyron built game to be modable by anyone. I was surprised, that he actually didn't say that anywhere. Most correct statement would be - VS was built in moding in mind. AI tools just took a step from other side and closed the gap ))) Being able to build something to play with, and share it with friends - is awesome!
  6. Thank you for opinion. I am sorry that my topic hit that hard. If you really-really want to fight against data-centers - you should support Ukrainian army. Our guys already burned some of them, and definitely will burn some more.
  7. Yes, it has been a long time there and it was PURE. And now author of it is using a modern tool for mundane job. I don't have a problem with it. But some people do(some of them already said that in this thread). I think I wouldn't reply again to you. I feel like I am wondering in circles
  8. Yes, only half get updates after release, and this is what has been a thing before a AI. Also, last couple of releases we had an x2 of author per release. In another words - every release there is a 50/50 split between veterans and newbies. It is kinda crazy.
  9. I am saying that AI has been used in their development. Are you following our discussion or just fool around? Have you take a look into repo of CarryOn?
  10. Just to make your struggle easier - lets take a look into top-20 mods from Mod DB sorted by most downloads. Check Carry On, Alchemy and Pet AI please.
  11. I don't know how much trust you have to me, I am just a stranger in internet - do you own research. If you need an example of something in top 100 - take a look into Forgotten Armory. Author give a good description for what exactly AI used in his mod. However, I see that you mix thing a little and confuse everyone in thread. When people say "AI mods are bad" they could mean two very different things. One is about ethic of use it. Something like "killing animals is bad". Argument can vary a lot - from "not at all" to "only if it is for looking into VS source code". Because its ethic - you can't argue other side quickly by any rational or emotional argument. For change in ethic - context is something that needs changes first. Another is about quality of such mods. Something like - "if it made with AI - it's sheet". This argument is subjective - people do like different things, and while it would be obvious for you that some mod is sheet - other people could find some fun in using it. Also, it is hard to measure. This is why I was looking into something that could be measured - support of mod over time. And I found out that if you are looking for something that wouldn't be abandoned in the middle of journey - label "made with AI" is almost the goods as label "made without AI". And much more powerful label - is a GitHub repository and couple of other mod in author portfolio that survived at least one game release. That it. Also, I find out that if you want to keep your mod folder clean with only "hand-written" mods - you do need to heavily rely on "no ai" statement, because there are some that use AI and don'ts state it, and adoption is growing.
  12. As a professional programmer I found it very annoying when people talk about code as art, learn by yourself, but can't learn git and GitHub. It is irresponsible in regards for community to do such things, and we tolerate it as nothing. Also, you trust that mod code doesn't do nothing malicious to you in case of closed sources mods is kinda brave. Code need to be published. Even if its art.
  13. Do I? Printing with printing press something handwritten is very expensive. And this is why people moves from fixed stamps to a variable ones. And you will not get the same quality of each page for free either - there is a lot of variables in printing, it is a complicated work. With AI variability of output is just a quirk that you need to take care of if you are looking for consistency of result. Is it complicated? Kinda. Is it impossible? No. Claim about stealing is an very aggravated to the degree where it misleads from what is actually happened. AI doesn't steal anything from authors. Stealing and plagiarism were a things before AI. And when they are done now by people with AI assistance - it still as disgusting as done by people without it. Let me show you some example. Did you find below image beautiful? If so, would you say that I "stole it" If I tell you that it is enough to know something like "z^2 + C" to be able to recreate it, and a dozen of other not less beautiful images? I don't think so. AI doesn't literally steal code or art from somebody. You can't find anything "stolen" in their pocket. The real issue is that to create something like "z^2+C" to produce code you need to have a lot of examples of code. And here comes real questions - how to honor authors of that code? How to check that their licenses wasn't broken by such use etc. It is more complicated then stealing.
  14. And AI does exactly the same. It makes it cheaper to produce. And you lose something in process. For printing press we lost a craft of handwritten text and custom made covers. BUt gain a literacy. For e-books we lost a paper smell and touch, but gain an library in your pocket. If code is art - should we ban all who doesn't publish it?
  15. This is your romanization of ModDB. Banner says: It's all about distribution.
  16. Nice comparison, and very powerful. AI still need to prove that it worth it, imho. I have not that strong, but more natural. Most of computers still runs on machine code these days. This is the same concept that was like from the start. And first programmers wrote their programs in machine code. They knew it and were very fluent in it. But that wasn't productive. So, people invent assembler. Was it good on the start? Hell, no! True professional could write much more better program that use less memory and would be quicker. But it would take time. And people started using assembler a lot. Do they needed machine code knowledge? Not, that much actually. They need to know boundaries of assembler usage: risk, use-cases, weak spots. The more assembler develops - the less machine code knowledge needed, and more of assembler knowledge. And then people invented C. And story repeated itself. And then people invented C++. And then Java. I don't know what Assembler-people feel about C++ people, but still remember what C++ people feel about Java-people. That was hate and contempt. Now we have AI. It still develops(and quite fasts), but still not in it final form. Yea, you would better to know language it writes program too. However, you should more focus on boundaries of AI usage: risks, use-cases and weak spots, imho.
  17. I said top 100. We are not in "top-10" world yet.
  18. Mine issues with AI are ethical too. However they are in a different areas and with some nuances. Don't want to push you into my point of view, just want to share. My biggest concern is when people detach them self from end result and doesn't take a responsibility for it. It's not a new problem under the sun, but AI amplifies it a lot. Next would be an translation. If I am a native speaker in one, and know another, and I am doing translation with AI, and I review it, and checking it against dictionary etc - caring about quality and proper language representation - for its fine. Other cases - nah. Art would be a gray zone for me. But it just because I just unliterate in art. If all previous century great people had a dispute on what is art and what is not I didn't see myself in comfort position for judgment. I wouldn't do myself personally some things, but wouldn't tell others what they should do. Let me give you an example. In my country Ai generated music is quite popular actually, like a lot. And it could be a bad slop music, it could be a song of a quality, or even an experiment like a band who make song by hands and mixing in an generated woman vocal in their singing. And I didn't see me telling these last guys to shut up. And code is not that big a question for me. Models trained on code of authors who explicitly prohibits such usage - non ethical, shouldn't be used. I wish all companies would be a transparent on this issue. However, using open sourced code for training seems fine for me. It seems fine for me, because I learned to code myself the same way. I've read the books where author were showing me how to code. I've read documentation where authors were showing me how to code. I've read source to check how to use libraries, and I was learning from that. Most of the written code had an requirements to be readable, so it mean to be read by others to understand it, to extend it, to support. In any commercial project that I was a part of - I was reading how it written and was writing similar to that. And whenever I got an opportunity to show other people how to write - I've showed them and was happy when they followed my path. Being somebody who always sells it works for other(I am still the author of that code, but not an owner in any means) - I am more then happy when my ideas lives even without me. This is why I see code writing with AI like something rather ethical, than not. I have an issue with vibe-coders as too, but it is again formulated in terms of responsibility, not a code understanding. Using AI for coding makes you not a coder, but a team leader(where you lead an AI), and I've seen a lot of failed leaders, and vibe-coders for me looks the same.
  19. Do I understand you right, that this is more of a ethical argument, than anything else? For me ethic is relative, so I just want to understand how you see it better. In mode creation I see three parts that could be ethical questioned - C# code, assets and translation. Do you find unacceptable usage of AI to create all of that? Or its more about assets? Would it be acceptable for you to use AI to create an browser extension that will filter out AI-created mods? And what do you mean by "have no respect for the craft"? Is it about coding? Art? Game design? Mods like "an infinite storage within your cattail baskets" - have respect to craft or not? Regarding donations - it is an interesting point of view, which I overlooked. So, here is a snippet of data: Share of mods and authors with a donation link (Ko-fi, Patreon, Buy Me a Coffee, PayPal and similar, found in the description and the mod's URL fields) ┌─────────────────────┬────────────────┬──────────────┬──────────────────┬────────────────┐ │ Cohort │ AI mods │ AI authors │ Non-AI mods │ Non-AI authors │ ├─────────────────────┼────────────────┼──────────────┼──────────────────┼────────────────┤ │ 1.20 │ │ │ 17.1% (188/1097) │ 9.0% (42/467) │ ├─────────────────────┼────────────────┼──────────────┼──────────────────┼────────────────┤ │ 1.21 (almost no AI) │ │ │ 10.6% (208/1962) │ 5.6% (49/882) │ ├─────────────────────┼────────────────┼──────────────┼──────────────────┼────────────────┤ │ 1.22 │ 23.0% (23/100) │ 14.8% (8/54) │ 8.8% (288/3275) │ 4.9% (69/1404) │ └─────────────────────┴────────────────┴──────────────┴──────────────────┴────────────────┘ 1.20 and 1.21 AI is empty due to very low numbers of mods and authors. Share of authors with a donation links in regard to expirience ┌───────────────┬────────────────┬─────────────────┬─────────────────┬──────────────┐ │ Author tenure │ 1.21 │ 1.22 │ 1.22 non-AI │ 1.22 AI │ ├───────────────┼────────────────┼─────────────────┼─────────────────┼──────────────┤ │ < 1 year │ 6.3% (73/1150) │ 3.7% (75/2043) │ 3.4% (67/1977) │ 12.1% (8/66) │ ├───────────────┼────────────────┼─────────────────┼─────────────────┼──────────────┤ │ 1–2 years │ 15.5% (57/367) │ 18.7% (108/577) │ 19.0% (107/563) │ │ ├───────────────┼────────────────┼─────────────────┼─────────────────┼──────────────┤ │ 2+ years │ 19.8% (95/479) │ 17.0% (128/755) │ 15.5% (114/735) │ │ └───────────────┴────────────────┴─────────────────┴─────────────────┴──────────────┘ How I read these numbers - newbies doesn't ask for donations usually. Established authors - do. This was a thing before AI, and a thing with AI as well. In 1.22 number of declared Ai usage is still relatively small, and within those mods/authors donation links are more often then among others. Difference between mods and authors due to some authors with good amount of mods, who collect donations.
  20. Data says a little bit different story. If we rank mods by downloads since 1.22 release - in top 100 there would be 7 of them that strait declare use of AI. And 5 that have some traces of AI usage: which could be an just a commit of contributor to fix something, or a help with some mundane task as code migration to new version of dot-net.
  21. Thank you for a warm welcome!
  22. Hi everyone! I have been playing Vintage Story for about a year and a half. For the first six months vanilla was enough for me, then I started adding mods for variety, and that meant reading the ModDB discussions. One topic regularly sets those threads on fire: AI. My impression (an impression of the tone, not a measurement) is that a lot of players feel AI slop has flooded ModDB. The claims I keep running into: there are way too many AI mods now; they are boring and low quality; they are poorly supported: bugs don't get fixed, code doesn't get updated; since making a mod with AI takes no effort, authors drop them quickly. I got curious which of these can be checked objectively and which are simply a matter of taste. Quality is a matter of taste: I can't measure it, so I won't pretend to. But "how many" and "how often are they abandoned" can be counted. ModDB has an API, so I counted everything: 9,291 mods, 50k+ releases, 3,411 authors (data of 02.11.2026). Guys, numbers are numbers. Where I say "I read this as" or "my guess", that is my interpretation, and you are free to disagree with it. All scripts, results and the method are on GitHub, so you can see how every number was made. So, welcome to the deep dive. 1. Are there really more mods? Yes. About three years ago, at 1.19 (January 2024), there were 1,323 mods. Now there are 9,291. That is x7 in 2.8 years, roughly +100% per year. If the economy grew like that, we would all own a house by now. Whatever else is true, this part of the feeling is right: ModDB has changed beyond recognition in three years. 2. Are mods dropped more often? No. To measure this I took every mod that was alive when a game version came out and checked whether it got an update within about half a year (195 days) after that stable release. Only mods with at least 50 downloads, so that experiments nobody installed don't skew the numbers. Share of mods updated: | game version | 1.15 | 1.16 | 1.17 | 1.18 | 1.19 | 1.20 | 1.21 | 1.22 | |---|---|---|---|---|---|---|---|---| | mods counted | 176 | 217 | 383 | 421 | 791 | 1113 | 1996 | 3289 | | updated | 57% | 69% | 51% | 59% | 48% | 58% | 50% | 52% | Roughly half of the mods survive each game release, and it has been like that from 1.15 to 1.22. 1.22 is no worse than before (52% vs 50% for 1.21, within the margin of error of about 2 points). So where does the feeling come from? I read it as arithmetic. 1.22 has three times as many mods as 1.20, so with the same share abandoned, you also see three times as many abandoned mods. The share is the same; there are just more of everything, including the abandoned ones. A caveat I will repeat later: "not updated" is not the same as "abandoned". Many content mods keep working without a single update. 3. The fate of a mod is decided early. Most updates happen right after a game release: about 8 of every 10 updates that happen within half a year come in the first month. Waiting from there to the end of the first quarter adds only 3 to 7 points. So a mod either gets its update quickly or, most likely, never. And the moment that loses a mod is usually the very first game release after it appeared. There is no fatigue with age: the more releases a mod has survived, the more likely it survives the next one (in 1.22: 47% for mods that have survived 0 releases, 56% for 1, 68% for 2, 70% for 3 or more). My reading: it is not that mods slowly fade away over the years. Most of them are lost at the first test, and the ones that pass it tend to stay. 4. The main factor is the author, not the tool. In both 1.21 and 1.22, authors with 2+ years of experience updated more of their mods than newcomers (under a year): 62-70% against 40-46%. Code mods are updated more often than content mods in every game version, by 9 to 29 points. My reading: the tool matters less than the person holding it. A veteran with a text editor beats a newcomer with anything. 5. Where does the flood come from? First we need to look somewhere else: at the players. Nobody outside the studio knows how many people play, so I used two proxies. First, downloads of 34 popular mods that have a release for every game version since 1.16: 1.20 got about 2.4 times more than 1.19, and then it stayed roughly flat through 1.22. Second, Twitch: streamed hours were about 3.4 times higher in the quarter 1.20 came out (January 2025), then a plateau, and a decline in 2026. So around 1.20 the game got a big wave of new players, and after that the audience stopped growing. 6. Where does the flood come from-2 In 2025 the number of new mods was x3.4 that of 2024, and the number of active authors x3.2. Mods per author did not change (about 1.7 new mods per active author per quarter). My reading: that is not AI, that is the audience from the previous part: new authors jumped in the same quarter as the players did, at 1.20. And then they kept arriving for another year and a half, even though the audience had stopped growing. My guess: the usual path "player, then user of mods, then author of own mods" simply takes time, and everyone walks it at their own pace. AI becomes visible from the second quarter of 2026: mods per active author rise from about 1.7 to about 2.0, which is roughly a quarter of the yearly growth, and newcomers more often start with code right away. The audience was flat by then, so this growth is not explained by the audience. How many AI mods are there? 549 mods declare AI use. That is 5.9% of all mods, about one in 17, made by 262 of 3,411 authors (7.7%). Not much for the amount of heat in those threads. But the share is growing fast: of the new mods, 1-4% until early 2026, 8% in Q2, 12% in Q3, and 24% in Q4 (the quarter is not over yet). If the trend holds, the share of all mods is still small only because the wave has barely started: it is just now approaching. Can we be missing some? Only 23% of mods (2,160) have a public repository. Looking for explicit traces of AI in them (agent instruction files, co-author lines in commits and so on) finds traces in 10% of repositories. But the same signs catch only 80 of the 178 mods that openly declare AI, so they see less than half. For mods created in 2026 with a repository, 19% show traces (169 of 871). A rough correction for the detection rate gives about 42-43% in this group, against 18% who declare. That is an estimate, not a fact: it is based on a biased group (authors with repositories are more engaged), and I can say nothing about mods without a repository. But I read it as: AI in 2026 is far more common than the declarations suggest. Which brings us to the question people actually care about: do these mods die more? 7. AI mods. At first glance AI mods look better: 71% updated against 52% for the rest (mods alive at 1.22). Then I checked when the declarations appeared. Many were added after the game release (active authors edit their pages, and some authors added the same sentence to many mods). I dated the declarations with Wayback Machine snapshots (for 8 of the 98 declaring mods the date could not be established, and they are left out). Mods that declared AI before the release: 56% updated (25 of 45 mods, 36 authors) vs 52% for mods with no declaration at that moment. No difference I can see: the gap is 3 points. The late declarers are at 96% (43 of 45 mods, 18 authors), and that is where the whole "advantage" came from. The same picture from other sides: - mods that declare "no AI" are updated even more often (80%, 12 of 15). But only 7 of those 15 declarations are older than the release; those 7 are at 71%, and all 5 late ones were updated. My reading: writing a declaration, in either direction, goes with being an active author, and adding it after an update is a sign of an active page. It is the effect of the author, not of AI; - hidden AI (traces in the repository dated before the release) are updated like other authors with repositories (75% vs 63%, tiny sample, within noise); - authors who declare AI come back with a new mod about as often as everyone else: roughly 4 in 10 against 3 in 10 among those who started in the first half of 2026. There are few of them, so all I can say is "not less often". So: the numbers do not show AI mods dying faster, neither the declared ones nor the hidden ones we can see. They do not show the opposite either: with 45 mods the difference from zero is not visible. 8. Conclusion. The feeling that AI mods are dropped more often is a vibe, and the data do not support it. Mods in general are dropped as much as always: about half at every game release, most often by newcomers and right away. The AI mods that have already faced a game release and declared AI in advance were updated about as often as anyone else. What changed is that there are many more mods and many more authors. That is a finding, not a verdict. Most AI mods appeared after 1.22 and have not faced a game release yet, so the data do not refute the vibe either. 1.23 will tell. Limits: The conclusion about AI rests on one game release (1.22) and 45 mods from 36 authors; "updated" is not "alive". The point will be made by a pre-registered check on 1.23: the method will be published before the data exists. Conflict of interest: I am a mod author myself and use AI assistants, including for this analysis. Objections and answers Nobody has objected yet, so I wrote the objections myself. Straw men are possible, so judge whether I tried hard enough. "Hidden AI mods sit among the 'non-AI' ones and spoil your numbers." Mixing in hidden AI does not flip the result by itself: if AI mods lived longer, the mix would pull "the rest" up, not down. It works only in a strong form: hidden AI mods are abandoned much more often than declared ones. Let me build that form. For AI mods to be truly worse than the rest, about 6 of every 100 "non-AI" mods would have to be hidden AI that is never updated (13 of 100 if they are updated 3 times in 10). That is not absurd, and I cannot rule it out: I find traces of AI in 2% of such mods with a repository, the traces catch less than half of the mods that openly declare AI, and among new mods the share is higher, about 28% after correction. What speaks against it: the hidden AI mods I can see are updated at 75%, not 0-30%, so the hidden ones I cannot see would have to behave the opposite way. So the honest result is this: the small gap between declared and the rest could be erased by hidden AI, and all I claim is "no visible difference". For the large gap (all declarers, 71% against 52%) the strong form needs 28 to 48 hidden AI mods in every 100 "non-AI", and that is not a serious estimate. "You are measuring declarations that were written later." Yes, that was my own first mistake, and it moved the result from 71% to 56%. The strong form: my dating is also imperfect, because Wayback Machine snapshots are sparse, so some of my "early" declarations may really be late. That would make the early ones look better than they are. I checked every disputed case by hand and left out the 8 mods I could not date. But it would take only about 10 wrong mods out of 45 to move 56% down to 44%, and I cannot prove that fewer than that are wrong. Traces in repositories are dated by commits, and anything dated after the release is excluded. "Authors who declare are just more diligent people." Agreed, and I cannot remove it: authors who write any declaration are not typical authors. Those who declare "no AI" update 80% of their mods against 52% for the silent ones, and that has nothing to do with AI. The strong form is that then no comparison of AI with the rest says anything about AI, only about the people. That is partly true: I cannot separate the tool from the person who holds it. What I can say is that the person matters more. The experience of the author (veterans against newcomers: +30 points in 1.21 and +16 in 1.22) and having a public repository (+8 to +16 points) move the numbers far more than the AI label does (+3 points, within the noise). "Mods of one author are not independent, prolific authors dominate." Right. I also count the share with each author weighted equally (1.22: 46% instead of 52%, the same ups and downs between versions), and I checked the AI result by authors: mods of one author were resampled together. No author dominates the AI group (45 mods by 36 authors), but the honest margin is wide. The gap between mods that declared AI before the release and the rest is +3 points, and it could be anything from 12 points worse to 19 points better. So I can rule out a big difference, not a small one. For all declarers against the rest the gap is +20 points, but that figure counts late declarations too. "Not updated is not abandoned, content mods do not need updates." True, which is why I split code from content. Code mods are updated more often than content mods in every game version. For AI there is a weak spot, and it is the strongest form of this objection: AI content mods are at 18% against 44% for other content mods. If that were real, it would be the biggest signal in this whole post: AI content mods dying more. But there are only 11 of them, from 8 authors, and the margin runs from 46 points worse to 10 points better, so I can neither confirm nor exclude it. AI code mods are at 70% against 61% for other code mods, and that gap is within the noise. Content is where 1.23 should be watched first. "AI mods are too young, give them time." Yes, and this is the main limit. The strong form: most AI mods appeared after 1.22 (by current declarations they are 8%, 12% and 24% of new mods in the last three quarters) and have not faced a game release yet, so what I measured is only the first 45 AI mods, whose authors may be unlike the rest. The only thing I can add is how new mods did in their first months: 56% updated at 30 days and 63% at 90 days for declared AI against 47% and 51% for the rest (61% and 76% for "no AI"). These use current declarations, so the AI figures lean upward. None of that is a verdict on what happens at a game release, and 1.23 is where this will be settled. "You only see mods that still exist." True. Deleted mods are not in the data. If AI mods are deleted more often than others, I undercount their abandonment, and I cannot check it. "What about quality?" I do not measure it, and I am not going to argue about it. Everything above is about how many and how often they are dropped.
×
×
  • Create New...

Important Information

We have placed cookies on your device to help make this website better. You can adjust your cookie settings, otherwise we'll assume you're okay to continue.