Very soon protesting the use of LLMs is like going to be like protesting the advent of the television. There is no stopping it. We should endeavour to ensure it is used ethically rather than becoming puritanical about its use.
From the article:
Danilov posited that the mistake was either the work of a “careless translator taking shortcuts”, or it was “done by someone on the dev/publisher side who couldn’t be arsed sending last-minute missing lines for translation and decided to throw them in a random LLM without oversight”.
Handong Ryu, who handled the Korean translation for the game, replied: "I was responsible for translating the vast majority of the Korean version of The Alters. Unfortunately, the same issue exists in the Korean version as well, which makes it more likely that the second scenario you mentioned is closer to the truth.
Sounds like this text was either added late in development or simply overlooked until after the last set of translation work had been completed, so the devs decided to let an LLM do it rather than getting billed for another batch of localisation.
Very dumb, especially as this puts them in direct violation of the Steam AI disclosure policy, but given the context I guess they figured no one would notice.
They could have used Google translate for these short last minute additions, and not a single fuck would probably notice. I hate this stupid overconfidence in AI.
Google Translate also uses an AI model.
When people are talking about dissatisfaction with AI usage, in this context they specifically mean Llama and GenAI. Google translate may use LLMs as part of their translation model, but it doesn’t make up the whole pipeline and will work completely differently than copy pasting some text into an LLM and telling it to translate something.
It probably uses a GPT of some sort at this point, tbh. There is no reason whatsoever using Google’s ML translation or ChatGPT’s ML translation should make any bit of difference to people who are actually upset over this if they have given any thought whatsoever to their concerns.
I have done translations and even for my own language I often use an LLM. It’s the one thing they are actually amazing at. It’s also probably not about “anybody noticing”. It can very much be a single developer doing it on their own ChatGPT account and the QA didn’t notice it.
I really don’t care about this stuff though. The AI label should be for gen AI and not revising some text or translation imo.
The issue becomes: if they used Gen AI to create background filler text, can we be 100% certain it wasn’t used elsewhere in the production of the game?
If they really wanted background text, they should have paid someone to write some, or use Lorem Ipsum if they didn’t want to spend the money and no one would have cared.
Look at the salty downvotes
“like it or not, gen AI is becoming an invaluable tool for developers”…
…who wish to take a dump on their work.
It will be used as a tool in pre-production and early stages of asset creation and no one will notice afterwards.
You’re expecting it to be used responsibly when we ourselves in general are very lacking in that department.
This here is a very good example of the actual use that will happen. A rush job to meet unrealistic deadlines. And that’s what will happen as is the norm.
We don‘t know the cause in this case. Not replacing placeholder assets was a common mistake even before ai tools.
That’s kinda the problem. We’re already careless with the things we do ourselves. It can’t be helped, nobody’s perfect. But once we start delegating tasks, we lose the direct experience. Priorities shift, attention moves to something else and the chance of carelessness rises because it’s no longer a problem we have to concern ourselves with.
Meanwhile, the LLM “learns”. What it “learns”, nobody knows because it does so mechanically. There’s zero understanding.
It keeps “learning” every time it’s fed something, so you don’t have a static program that does what it’s told. Instead it’s a “living” program that applies what it “learns”. And that makes it unpredictable in the long run.This turns the user into a glorified middle manager who has to hover over their employee and make sure they did their job as they should have. And how many middle managers do you know with that kind of dedication, that isn’t spiteful at its core?
The push against this is that the people depending on it to do the work become less dependable themselves. And unless you’re an independent developer without a profit driven publisher breathing down your neck, this will be used in all the wrong ways as a standard instead of it being the exception.
I don’t think it’s important where the placeholder assets come from, or that mistakes will be more common when someone used gen AI instead of non-licensed stock image from a web search.
You’re right. It’s an opinion and only as important as the one having the opinion decides it to be.
According to the article as cited in this comment, we do know the reason and a rush job to meet a deadline is precisely why.
I wouldn’t say „precisely“ as those are (plausible) speculations.
Super weird take, honestly. Procedurally generated content gets no hate, despite it being janky dogshit, too.
EDIT: lol your downvotes don’t make your opinion more consistent
You must be young. proc gen used to get tons of hate in the 2010 and such era, gamers complained about devs being lazy and not being willing to actually make levels/worlds/dungeons/whatever. This complaint was of course inconsistently applied.
These days people mostly just got used to it as normal. In 10 or 20 years, I’d wager the same will be true of gen ai.
I’m not and it’s always been consistently praised.
I will concede that we have lived different experiences.
Totally valid, mutually conceded. I’d bet we can agree that the current climate of games generally praises procedurally generated content, regardless of how we experienced its history.
Agreed.
There’s more than one argument against generative AI being used in games, and they don’t all apply to proc gen content. It’s an apples to oranges comparison in most cases.
And yet you couldn’t describe one aspect of the differences 🤔
Ita because you are still putting in the work to license or produce the individual parts used in procedural generation rather than using people’s work without pay or permpermission.
Love and hate are subjective opinions, so of course they’re unfair.
And so are upvotes/downvotes.
Edit to clarify: what I meant was, if you don’t understand why procedural generation is acceptable, and generative AI is not, you are not qualified to have an opinion on the subject. Leaving the original text for context.
If you don’t know the difference between procedural generation and generative AI, you are not qualified to have an opinion on the subject
While your statement is objectively true, it does not pertain to the comment you replied to. Read it again, they were making a comparison. They did not claim that the two things were identical.
I feel like it does. theunknownmuncher thinks it’s somehow inconsistent to be against generative AI while being ok with procedural generation, which implies that they think they’re equivalent in some way. As if the reason people don’t like generative AI is because it makes bad games.
Edit: throughout this discussion, my opinion has evolved somewhat. Procedural generation is fine, because it only uses things created by the developer, and it will necessarily generate a better product than a generative AI, because the developer is the one who tunes it. An AI will generate any text that might fit within the genre, with no consideration for what’s canon to the work it’s being inserted in.
both are used to produce more content with less effort. There’s your equivalence.
What would actually add value to the conversation is discussing why a particular criticism of one may or may not apply to the other.
I actually disagree with the original premise, and explained why in another comment.
both are used to produce more content with less effort. There’s your equivalence.
Bingo.
As if the reason people don’t like generative AI is because it makes bad games.
Nice, point proven. 😎 If it doesn’t make games bad, then the complaints are simply invalid and bandwagoning, and developers cannot be faulted for using it. LOL
Point not proven.
There are many reasons why people in general actively dislike generative ai. Many of those reasons have to do with the creation of the ai (including environmental damage and harm to artists, and more besides), and are applicable regardless of the quality of the end product.
Furthermore, using generative ai does tend to make the end product worse, regardless of what that product is. This does not mean that it is impossible to make good shit with ai, nor does it mean that ai only makes good shit. There’s nuance to the issue that is often ignored.
Furthermore again, there is bandwagonning happening in the hate of ai. However, just begause bandwagonning is a logical fallacy, does not automatically make the arguments wrong (see the fallacy fallacy).
Furthermore the third, developers absolutely can be held at fault for using generative ai. Valve demands ai use be disclosed, they didn’t comply, ipso facto, devs are at fault. However, not all fault is equal. The example being discussed in the original post is much less egregious than most in my opinion. It’s not like they ai generated the entire game asset by asset.
I had another point but already forgot what it was so I’ll leave it at that for now.
If it doesn’t make games bad, then the complaints are simply invalid and bandwagoning, and developers cannot be faulted for using it. LOL
“If slavery doesn’t harm the economy, then the complaints are simply invalid and bandwagoning, and plantation owners cannot be faulted for using them. LOL”
I know Lemmings have a lot of trouble reading, so I’ll get this out of the way now: no, I’m not saying that generative AI is slavery, nor am I saying they’re equivalent. I’m drawing one similarity to make a point. That’s called a simile. The point being, that one supposed criticism isn’t valid doesn’t mean that no criticisms are valid.
Sharing one thing in common does not make two things equivalent. You’re welcome to try again though
you demanded an equivalence. I gave you one. If you don’t like it then that’s a you problem.
Your previous comment proved my point, thanks
LOL care to educate us on why a statistical model is unacceptable while a procedural model (also statistical 🙃) is acceptable, then? 🤔 I’ll wait.
(reality: it’s a minor implementation detail and has no relevance to the user)
There’s a number of reasons, not least of which being that generative AI works by processing vast amounts of prior work (without their creators’ consent) to make a facsimile of it, while procedural generation only manipulates assets the developer creates. Procedural generation isn’t putting artists and writers out of business. Procedural generation isn’t making Idiocracy a reality, with fucking English majors unable to read Dickens without asking OpenAI to interpret the text for them. “They do similar things” doesn’t mean they’re equivalent. My point being, it’s not inconsistent to be okay with procedural generation and not okay with generative AI.
this is a fantastic point. well put.
If you think “AI” and a designed classic algrithm generating things are equivalent, no wonder you hail AI as good… because that is fucking clueless take.
It’s literally just implementation and they’re both statistical models, but 👍
If you disagree, explain how. I’ll wait
no wonder you hail AI as good
When, exactly, did I? I called them both janky dogshit, but simply pointed out the very real hypocrisy of supporting procedural generation while hating generative AI.
“You’re not actually supposed to read that text so this is not an issue.” Good job missing the point.
That’s kinda sloppy, mainly in the disclosure and translation department, but nothing that some updates won’t solve.
I’m happy to support these devs with original ideas, even while they use some AI in a non-intrusive way. They have done something more important for in my book, which is following regional pricing.
It’s stupid how several studios think charging my 3rd world country the same or even more than the US is a good idea. CDPR and bethesda think it’s ok, but It’s disgusting. I would rather support 11bit.
Damn, I was looking forward to playing this. Glad I read this first
Boycotting because someone made a mistake? Ok
Boycotting because they used generative AI to make their game instead of hiring writers. Even if this was the only part of the game they used it on (if you believe that, l have a bridge to seII you), I’m not going to give someone money if they couldn’t even be fucked to hire some sci-fi writer off of fiverr to write their fill text with it.
I personally know artists and writers who are having to get jobs at fucking Walmart because of this shit. I’ll be less irate about generative AI once we have universal basic income so that real artists can continue to generate real art alongside these soulless husks.
So they shouldn’t even use it as placeholder assets, instead of Lorem Ipsum?
Correct! If you’re not going to support artists and writers, the least you can do is not support the industry that’s actively destroying the fields of art and writing (on top of the myraid other problems with generative AI)
No artist or writer would have been hired for producing placeholder assets.
So do you just not play video games at all? Because the way you’ve just presented yourself you’re not against using AI in games, you’re against any use of AI.
How can you be sure that in any game AI wasn’t used to generate some sort of an internal document or asset that would never be in the final product but was integral to the creation of the final product? Clearly you don’t write every dev and ask if they use AI in any capacity, so what do you do?
My point is that I think you’re taking a stance where you’re unwilling to compromise on the use of AI, but only if you’re aware that AI was used.
How can you be sure that in any game AI wasn’t used to generate some sort of an internal document or asset that would never be in the final product but was integral to the creation of the final product?
I never claimed to be omniscient. I simply don’t support a company after I find out that they have unethical business practices. What are you not understanding about this?
Got it, you’re fine with AI usage as long as you don’t know about it.
