The primary use case for House House’s use of LLMs was to research, via chatbots, issues we encountered with third-party code. If you’re technically-minded, and curious, here’s an example. We needed to cull shadow casters by distance from the camera, but Unity URP only allows culling shadow casters by distance in camera forwards, so shadow casters behind the player were being rendered at an infinite distance. This made our game run 50% slower for around 20 seconds at sunrise and sunset. We researched the problem extensively, and we looked at the URP code ourselves, but we couldn’t find a way. Then, we were able to speak directly to an engineer at Unity, who asked around and ultimately concluded that what we wanted wasn’t possible. Finally, we posed the question to an LLM chatbot. After a few dead ends, it suggested a possible fix, and while we had to modify the suggestion extensively, it worked. This fix represented about 10 lines of code inside URP.
asking a LLM for a fix that added about ten lines of code after heavily modifying the LLM output is exactly how AI should be used. They exhausted all other options and still had to tinker with the LLM code to make it work. if that’s not real coding, idk what real coding is anymore
for those that don’t approve of this, why? just curious, no hate 🙂
There are endless threads to discuss in this, but two really seem lost in this debate:
-
The distinction between vibecoding, and lower levels of LLM tool usage.
-
How widespread the latter already is in software development.
Basically, if you’re a dev coding for money… you’re already interacting with text models, somehow. You have been. Everyone reading this post is using code touched by them, up and down the stack.
Hence I don’t think people realize how impractical the “purist anti AI” stance really is. It’s like asking devs to avoid anything touched by Microsoft or Facebook: a good idea on the surface, but how on Earth would you actually do it?
I think what gamers really hate is higher levels of vibecoding and tool usage to make slop. As they should. As well as support for authoritarian entities backing them. But this just isn’t a practical debate without discussing where to draw the line, rather than an absolutist “LLMs or no LLMs” take, as the latter is going to catch basically every dev on the planet.
You can just not use it personally?
In the modern internet and tech, get away from AI entirely, period. It’s being shoved in our faces everywhere, it’s copilot on your Windows 11 machine, it’s being used to develop Linux distros, most numbers you call for support force you through AI bots first.
-
Big walk might be the first must play group game to come out of the friend slop category. It was such an elevated experience by the end I was thrilled. The very last scene almost got me to cry and we were just logging off. Crazy.
No amount of this game could be made with AI that would be ultimately upsetting to me unless they were personally responsible for a meaningful billing to the worst LLM companies.
I have a lot of serious ethical concerns about AI and vibe coding. But I do recognize that there are legitimate use cases when used as a tool and not a crutch. And this one really don’t seem like the end of the world here.
My concern is that the takeaway from an incident like this is that next time, devs are just going to lie and not tell anyone they used AI. They didn’t have to say anything and no one would have ever known, but they’re being raked through the coals for responsible disclosure.
If I was a developer, I’d just lie.
If I was a developer, I’d just lie.
So it sounds like your “serious ethical concerns” aren’t quite so serious after all.
I think you are willfully misinterpreting me. AI is a nuanced issue, but I don’t have a problem with this type of use case. And these developers did the ethical thing in disclosing, but they’re in trouble for doing so.
My point is that excessive witch hunting gives developers strong incentive to white lie.
Here’s the statement from IGF jurors that’s related to this. (The link is likely a repost; I don’t think the person is affiliated with IGF.) This hardline, purist stance removes all nuance over when AI. The question shouldn’t be if AI should be used at all. It should be more about how it is used. Adopting this stance doesn’t mean developers are going to be suddenly honest. Some devs will be even without this stance, but many others will take time to hide or disguise their AI use. The part about coding especially bothers me, because AI can be very useful for debugging, but to them, based on this statement, any use of AI is the same as vibe coding a whole game.
It sounds fine. “Any AI use harms the environment and community creation. So we’ve decided to vote against it”
My biggest concern with this is that they still don’t have the required disclosure on their steam page. Can they really be trusted if they can’t even be better to follow Steam’s ToS?
They did follow the disclosure requirements. See the article.
Please provide a screenshot of their steam page that clearly state’s their AI usage.
Please read the article.
As for why House House is only confirming AI use in the game now, rather than at release, they say “We have not made any disclosure on Steam as the platform does not request disclosure of code or chatbot usage. We are posting this disclosure as it has become a matter of public interest.”
I had missed that, seems like this is a failure on Steam’s part
I am surprised the IGF jurors are having arguments about anything, I just assumed all these awards things were rigged for marketing.
They don’t have an Ai disclosure on the Steam page: https://store.steampowered.com/app/1478500/Big_Walk/ Isn’t this against Steams rules?
No, as per the article, the way they used AI does not require disclosure on the Steam page.
Where can I find the rules for Ai disclosure on Steam? I couldn’t find it.
It seems to mostly be in the developer portal, but here’s the news post about it.
This is the news from 2024, which I found too. I was looking for the actual rules a developer has to agree to. And the users need to know too, because we can report. Any news article talking about this has no link. It’s frustrating.
Quite frankly, I don’t think it’s on us. The devs have their forms to fill out, and they’ve disclosed this publicly of their own choice. If Valve found that usage to be in violation of their own policies, they’ve already got repercussions for this in place that House House would not want to be subject to.
Other developer information is online available for everyone too. So this is usually how Valve operates. I think I found it: https://partner.steamgames.com/doc/gettingstarted/contentsurvey The part about the Ai disclosure is exactly the same as in the news article. I was just wondering if there is more to it. So its only about generative content then.
As a developer, I really think the reactionary response to AI is muddying the waters. Their stated use-case seems perfectly reasonable, as their options were “shit optimization with even the unity devs clueless” or fixing the bug. AI makes a great tool for some applications. No, it should not be used to vibecode entire apps alone, but it does have the power to search and aggregate sources far, far better than humans can. It will likely take some tweaking afterwards, because their output must not be trusted, but using AI as a research tool for solving esoteric dev problems seems perfectly reasonable to me. I would much prefer we not punish the people that are fully disclosing their (very limited) use, when many of the competitors simply aren’t disclosing.
This is exactly why barely anyone discloses AI use anymore. I use AI sometimes, my colleagues use AI sometimes, everyone I know in the software industry uses Claude in some capacity. But we don’t talk about it in public because it’s become so toxic that you might as well replace a simple AI disclosure with “please witch hunt me and call months of my work vibecoded slop even if it clearly isn’t”
Most people are over AI being used in code but some people aren’t ready for it yet. I give it two years until it’s basically normalized.
There are inherent problems with Ai, which makes any use of them regardless of what it is, problematic. Not everyone cares about the problems. Things like not respecting the license of training data, and other ethical reasons like Ai companies being responsible for the RAM and graphics cards shortage / price hikes.
If you ignore these problems, then we can have a talk about responsible usage and not doing slop. Regardless of what you think about Ai usage, I hope you at least understand and respect the opinion of those people too. Because its grounded with (good) reasons.
I would much prefer we not punish the people that are fully disclosing their (very limited) use, when many of the competitors simply aren’t disclosing.
They did not disclose the Ai usage on the Steam page.
Not all LLM trainers are so caustic to the world. A few use open datasets for the vast majority of their training, like Nvidia’s Nemotron series or a few Olmo derivates.
Many LLms are also trained on shoestring budgets, with a relatively negligible environmental impact.
But are also other huge ethical concerns, like entities behind them supporting censorship, fascism and such. The erosion of the world’s mental aptitude from how it’s presented. Security, labor, hacking and spamming, and so on.
…This is the thing with discussing “AI.”
It can be as nuanced as things get, but few care. Most seem to want things in simple absolutist terms, and they aren’t interested in any nuance, especially the gamers on either side of this debate.
Oh, absolutely. However, I also recognize that we live in an era where it’s impossible to have completely ethical consumption, and I suspect many of those most vocal against any AI use are generally being more selective towards this one topic. I personally don’t begrudge the (disclosed) use of the tech for applications where it is genuinely the best option available. The issues around training data and environmental concerns (even ignoring the effects of the economic bubble) are real and valid.
For the record, I do not use AI or support AI use, but I recognize that things are shifting and it will likely become a permanent tool. Hopefully the ethical issues will be ironed out before it’s too entrenched. I suspect once regulation is in place and the bubble pops, what survives will be much more sane.
I personally believe it is impossible to have no ethical issues with Ai. Despite this fact, I experiment with local Ai chat tools. I used it when coding offline, to lookup ask some questions I would have asked a real user in example code snippets if it finds any issues I couldn’t see. Or how something is achieved usually, when I’m new to some stuff. That’s basically the extend of my usage in projects so far. Even if I wanted to slop more, I couldn’t, because my hardware is weak for this task (still don’t want to).
I like it for extracting text from images and translate it, generally speaking translation with Ai is much better, I play games where I know Ai was used to create it. As you see, I’m not completely against it.
But this is me personally, and I still have these ethical thoughts and it feels dirty.
Agree. I have been tinkering with local solutions, but nothing quite fits my use case. I have messed around with the latest DeepSeek model, and it seems quite powerful. Lower training cost and energy footprint is a plus, but still is fed by the same scraped content. I suspect the ability to write syntactically correct code and expand pseudocode will be the most useful feature of AI that persists post-bubble.
I’d challenge the AI purists by asking where they draw the line. What if your developer uses AI to create every single texture and mesh? What if the entire codebase was built with AI assistance? What if the developer used AI to fix one single gremlin of a bug? What if they create all of their regression tests with AI? What if the developer needed a refresher on some basic C code syntax and got an AI summary from Google - where AI never touched the code?
Are all of these really equally heinous?
While I wouldn’t consider myself an AI purist I do have different views when it comes to using AI for coding compared to using it for art or music.
I feel that expecting a game to be completely AI free when it comes to coding is pretty much an impossible task. Now matter how much you try at some point a developer will have to use a tool or library or whatever that were developed with the use of AI. Expecting them to audit everything they use would mean spending more time on that than coding the actual game or spending time developing non-AI alternatives. The genie is out of the bottle for coding and there’s no putting it back.
Now when it comes to art, music and music I personally am less accepting. AI has been trained on the art of millions of people around the world without their permission or offering them any compensation. That is unacceptable to me. It also feels very weird that people are trying to replace art with something generated by a machine, it feels soulless, each piece of human art no matter how simple has some meaning to the artist. AI just mashes it together with no rhyme or reason.
Now some people are going to say that I’m a hypocrite and coding AI has also been trained on the code of millions of people without them getting compensation or asking for approval. While that is true I think that for coding it’s different just for the simple fact that at the end of the day it’s just a set of instructions. There is code that could be considered art? I would say there definitely is but the vast majority of it is just telling the computer to do a thing in just slightly different ways. This is just how I see it and no doubt people are going to disagree.
While I agree with the main message, I’ll be one to say you are a hypocrite. Code doesn’t have to be art for the use without consent to be wrong. Arguing that way is the same as when people say that using AI to generate images is fine because they have the idea but not the skills to execute it.
Either way you are diminishing someone’s else work because you find this tool useful to achieve the results you want.
I draw the line at using it personally. I can’t control everyone else using it but I’ll gladly prioritize the work of someone who put the sweat in over one that simply prompted a model.
To me the AI use has become tantamount to endorsing and accepting the problems it is creating. Once we solve the problems it has created, I’ll feel fine using it again.
When it comes to the harms of these systems, that has a lot more to do with how these companies allow use of these tools…
The real harm is in these tools being pushed as learning tools, medical advice, companionship, etc. That they can even give medical advice seems like something Doctor’s Unions should be suing over, I can’t do it so why can the AI’s? Using these in learning situations is demonstrably doing the opposite, inducing cognitive offloading and dependency. The companionship thing is probably the most dangerous facet, it’s causing genuine psychosis in some people and honestly I feel that is by design.
None of this has to be for these tools to be used for coding. They represent a regulatory failure (complicity?) to constrain the dangerous uses.
Whereas I work with a small studio (<10 ppl) as the only programmer, asking Claude for bug hunting in regex’s and shader code has saved my behind a few times.
I’d argue there is a responsible way to use these tools. Letting an agent run amok in my project folder is a hard no; I ask about specific issues, providing generalized code, and insert and test the code by hand to ensure reliability. The amount of code inserted by an AI is zero, but the code written by AI would be <1% of my codebase.
I might use the tools for coding but I advocate locally to keep them out of schools, and support all efforts to limit the harms. You can use a tool for a useful purpose (hammer for a nail), while still restricting that tool’s usage (wielding a hammer to threaten people). So I disagree that using these tools is tantamount endorsing them for all purposes.
Frankly I’ve never paid any of them a dime, so I’m actually driving up their costs and hastening the decline 🤷
Nah, too many friends of mine have been fired because of “AI”. It’s the mothern cathedral, what the powerful are using to make us dependent on them. I can argue than even when you didn’t pay them a dime, you still gave them training material for the next model; furthering their goals ever so slightly.
I’m taking a stance that looks harsh on the outset, but it’s basically the only one that doesn’t further the massive theft of our body of work onto the hands of 5 oligarchs.
it’s disappointing but i get it. these tools are meant to be alluring and addictive. i’d be lying if i said i didn’t give in to the temptation a few times, as much as i hate to admit it.
i still think it should be disqualifying for a game awards ceremony. if not, i worry that this could be used as a foot in the door by slopbros to further normalize ai use in games (and, in turn, elsewhere)
I’m not sure this is a great policy. You can tell the difference between using ai for debugging and a whole game made ai slop. Nobody in that second category will be just sneaking by unnoticed.
But using ai in the final stretch to squash some bugs, that’s pretty hard to notice from the outside. If you outlaw the use of ai entirely, what do you get? People still use it for the little things, but don’t admit to it. Is that better?
The primary use case for House House’s use of LLMs was to research, via chatbots, issues we encountered with third-party code. If you’re technically-minded, and curious, here’s an example. We needed to cull shadow casters by distance from the camera, but Unity URP only allows culling shadow casters by distance in camera forwards, so shadow casters behind the player were being rendered at an infinite distance. This made our game run 50% slower for around 20 seconds at sunrise and sunset. We researched the problem extensively, and we looked at the URP code ourselves, but we couldn’t find a way. Then, we were able to speak directly to an engineer at Unity, who asked around and ultimately concluded that what we wanted wasn’t possible. Finally, we posed the question to an LLM chatbot. After a few dead ends, it suggested a possible fix, and while we had to modify the suggestion extensively, it worked. This fix represented about 10 lines of code inside URP.
“Temptation” is one way to put it, I suppose, when the human experts tell you it’s impossible. This award’s criteria was already decided, and they’re entitled to their definition, but this is just information that was surfaced to them rather than work done for them.
Unity support devs saying things are impossible is like in fight club when they always say you can’t join.
They don’t really mean it. They just have other things to do.
They used AI for this??? Its like the most basic game ever.
It’s really not. A multiplayer open-world game is inherently complex to develop, even with modern engines.
It is. If they were gonna use AI, they could have at least used it on the voice chat system, as it was fucking shit.
…to make it even more shit, I take it?
Fans are out en masse!
Can’t wait to see your full remake of the game since it’s so basic
Could do if I had time!
You’ve got plenty of time. Look, you just saved a whole minute by not reading the article before you commented.
You can make an open-world, online game in a day in Godot or Unity by hand with no AI. It is not difficult anymore to make simple online games. You’re just saying things and hoping they’re true.
Trivializing other people’s professions is what AI-bros love to do the most. “How hard can it be to scan grocery items” ahh comment.
I have worked on an online Unity game, and it was not made in a day. It even came from a good template of a starting point, but everything you build on top of it is extra complexity that needs to be tested, debugged, and fixed.
No, I’m saying things I know to be true, because developing 3D multiuser apps in game engines is a large part of my job (and has been since well before AI).
You can throw a couple GameObjects into a scene and synchronize them pretty quickly, but that does not a multiplayer game make. Every single script you write will be more complex when you need to keep multiplayer in mind and testing anything you’ve written becomes much more involved. This is especially true for a large Open World like Big Walk’s.
Yeah, the tools in these engines are really good now and do all of the heavy lifting for you.
Perfect for ai then
All professional software development is using AI. This is normal and to be expected. Code AI contributions are rightfully handled differently in Steam’s AI disclosure.
Note how pretty much every major piece of software has gotten less reliable and less user friendly in the last few years. There might be a connection…
I forgot the golden role of “ai bad”.
AI is a tool. A useful tool for software development. But with any tool, it’s possible to misuse it. With AI it’s easy to abuse it and get it to write everything for you and not think about the output. That usually ends up with bad code. You can also use it for debugging or code review or giving a solution to a tough technical problem that can be used as a starting point. That’s how these devs used it. That is a good way to use the tool.
Just because corporations overuse AI in software development doesn’t mean the tool is bad.
People have complained about this for as long as computers existed. It’s not a new development and not exclusive to LLM usage.
People have complained about this for as long as computers existed.
That. Doesn’t. Mean. It. Hasn’t. Gotten. Worse.
Actually comment for yourself instead of parroting thoughtless bullshit.
That wasn’t the implication. The comment above mine could be read in a way that suggests, declining software and systems quality has suddenly emerged from LLM usage. I was trying to highlight, that this has forever been thrown around as an anecdote. I hope, that someone reading my comment would think more thoroughly about their experience with software over the years and how their own perceptions and bias might have changed through that time.
There is, but it’s not AI alone. It’s mostly the C-Suite trying to drown us in it.
Developers House House told Aftermath via email that “LLMs were used sparingly by House House to help work through difficult technical challenges and bugs”, and that they “estimate that well under 0.1% of Big Walk’s code had anything to do with this usage”.
Meh, if that were enough to exclude a game from GotY awards, you’d end up with a very short list of mostly unknown games, half of which probably did use LLMs for those sorts of tasks and are just lying about it.
And honestly, if AI-debugging can replace having to spend days of horrible crunchtime fixing those bugs, I’m all for it! That’s what AI should do, making people’s lives easier and not replacing their creativity.
Honestly this ordeal is actually bigger than people think. because it stems from the IGF written statement from 19 of their 50 or so jurors.
IGF has a decision to make here and it isn’t a easy one. They need to either choose whether the jurors are allowed to hold that opinion, which marks them as a pure bias source which is not good for a jury that is meant to decide the final choice on your listings; say “yea that’s fine” and basically throw out any form of credibility that IGF has in it’s voting process like how TGA has been for years, or just not put 19 of their jurors on any category that contains a game that has AI in it (which isn’t super difficult as a category only has 7-10 jurors)
Personally I think that IGF is in a damned if you do, damned if you don’t type situation here cause either one of them is going to screw their publicity.
“Difficult technical challenges and bugs” is what I find AI is the worst at in my experience.
You conveniently dropped “to help work through” here, so now it implies they let AI do the work.
I think the example cases they’ve provided are pretty good example of where an LLM can actually be useful. For example, if you need to find and fix a bug in third-party code, it can be an absolute pain, as just getting a basic understanding of an unknown codebase by yourself could take days or weeks. An LLM can analyze the code in seconds and if not fix the bug, at the very least point out locations where to look for it. It’s not necessarily about telling the LLM to overcome the technical challenge by itself, but it can definitely assist in it.
“Everyone who says they hate AI is actually using it and embarrassed” is THE biggest cope I’ve ever read.
I guess so, but I’m not sure where you’ve read it because I sure haven’t written it.
hooray, insane witch hunts and even more insane mental gymnastics galore!
(Big Walk is awesome and you should play it)
