How we cover AI-assisted projects - GBAtemp community check-in

GBAtemp Banner.png

It’s about that time of year again where I check in with the community to see if we can get the front page aligned with what people are interested in knowing and what people are interested in talking about. As the title might have given away, there’s one topic more hot than most at the moment, and that’s AI.

I am a C++ developer by trade, and my place of employ has just started to roll out Claude licences for those interested, so the topic of AI assisted programming is very much on my mind. I’m not just going to go out there and say “AI bad”. I don’t necessarily think it’s a productive conversation to have. The reality is that we’re going to see more projects that make use of it in some form, and some are likely to warrant attention.

Going forwards, I think it’s important that we highlight when projects we do feature make use of AI, and to what extent if such information is available. My hope is that this will strike a reasonable balance of being able to share projects that might look interesting, and giving people who want to avoid AI related projects a simple indicator.

Obviously not every project is going to be so open with their usage, and we have no intention of starting witch hunts by deep diving code and looking for em dashes in the comments. The aim here is to make whatever public information there is accessible for people reading here, and just make things a touch clearer for the people who are looking for that kind of disclaimer.

With that aside though, I am interested to know what the general consensus is among the user base. Are you open to playing ports assisted by AI if it’s the only port available, or happens to offer something unique? Do you draw the line at vibe coding, or do you avoid any project that has Claude listed as a contributor? We sadly don’t have access to polls on the front page any more, but I will make sure to keep up with any replies as they come in.
 
Those who say they don't mind if it's fully AI and virus free forget that the AI makes mistakes so a human should be there to review and verify that it's passed the test.

Personally, I don't mind AI as long as it's mentioned and there's a human dev to check it's good to go.
 
Those who say they don't mind if it's fully AI and virus free forget that the AI makes mistakes so a human should be there to review and verify that it's passed the test.

Personally, I don't mind AI as long as it's mentioned and there's a human dev to check it's good to go.
AI doesn’t work that way

I can guarantee you that no one is saying “hey ai make this project for me, make no mistakes and upload it” it makes enough mistakes where if you don’t add that human touch you are either fucked or burned 6x the tokens you would otherwise to get your results

That is just a waste and time for credits for everyone involved
 
  • Like
Reactions: bonkmaykr
I already posted most of my thoughts on AI projects in another thread, so look there. But more related to your question:

I do think it would be to the benefit of everyone if projects were transparent about their development history and this was reported on. The witch hunting thing is not our problem. Those people are the loud, unimportant minority, and they can gladly make an ass of themselves.

GBATemp should not however decide what projects get reported on or posted about based on this history. As we can see from this conversation, this means different things to everybody.

Ai has been in games for many years already technically, NPC's are essentially Ai to some form.
That is a completely different thing for a completely different purpose.
 
Last edited by bonkmaykr,
"Hate the player, not the game."

If someone's just throwing crap at the wall using AI and couldn't tell you how any of it works if taken to task then that's the problem, not the fact they used AI to build it.

I rate the competency of both what was made and what the developer is capable of answering for when issues are levied. That's the real test of whether or not something is "AI slop", at least in the realm of coding.
 
the only people complaining about this haven’t made or shared anything publically themselves. So without disclosure they have zero way of knowing. All you’re doing is giving them the easy target to be toxic

If this becomes a thing I def won’t be disclosing anything and will take any warnings/bans that come with it. It’s no one’s business or exactly fair to dismiss the whole
Project no matter how little or a lot AI may have been used
Outright incorrect, and such a baffling claim to make. I try to keep AI out of my own projects, because otherwise what am I doing? And all the same, I try to avoid AI-generated content to the best of my ability, including code. If you feel that you need deliberately to be deceptive about your work practices to get people to use it, that reflects way worse on you than just using AI. But, it's your reputation to build, so do with it as you will.

As for the original topic, I think AI projects should definitely be disclosed, but reported on when significant. We don't need to flood the front page with the hundreds crummy RexGlue recomp that've been shat out over the last few months. As a community, I'd rather we focus on what people are actually doing, not what they've asked computers to do for them.
 
Honestly I don't think it matters in the gaming universe much. If there is a port of sorts without alternative I'm REALLY happy about it.
If it's not a hidden software that talks to the internet I don't even mind it beeing completely vibe-coded since maintanance is not that important if it's just a game (and runs).
 
Outright incorrect, and such a baffling claim to make. I try to keep AI out of my own projects, because otherwise what am I doing? And all the same, I try to avoid AI-generated content to the best of my ability, including code. If you feel that you need deliberately to be deceptive about your work practices to get people to use it, that reflects way worse on you than just using AI. But, it's your reputation to build, so do with it as you will.

As for the original topic, I think AI projects should definitely be disclosed, but reported on when significant. We don't need to flood the front page with the hundreds crummy RexGlue recomp that've been shat out over the last few months. As a community, I'd rather we focus on what people are actually doing, not what they've asked computers to do for them.
Deceptive isn’t the word I would use and you are correct I shouldn’t have said everyone, but the vast majority of folks havnt contributed back themselves besides complaining. That I whole heartedly think is true but you are correct. I shouldn’t not have said anyone

But with your other paragraph most things wouldn’t be shown off anymore following that logic or be allowed

Look at the renaissance of switch and 3ds projects that havnt been made before or other devs claimed impossible before these came out

And like my own personal example, most big devs and companies use it now. A lot of the ps5 kernel and userland entry points would not exist currently without it

There is no way to draw that line accurately without causing a witch hunt or alienating people

And it’s not going to stop them either we are just going to lose members and releases here
 
  • Like
Reactions: bonkmaykr
AI =/= LLM
You're 99% correct. Language evolves over time. Just like how "whoa" is the correct spelling, enough people spelled it "woah" for long enough that eventually, they just started adapting it as an alternate spelling. LLMs and machine learning aren't true AI, but enough people are calling them that nowadays, that the only way to distinguish real AI is to call them AGI or ASI.

For the record, I'm a stubborn pedant, so I'll always refer to what's currently available as LLMs or ML no matter how the language evolves, just like how I'll always spell it "whoa" and not "woah". "Wh" is a sound. I'm not going to start spelling "what" as "waht" or "who" as "woh" either.

On topic: As others have said, transparency is paramount.
 
I would very much appreciate a disclaimer/notice that the project uses AI in any form, when possible. I personally don't have any interest in engaging with those kinds of projects, and nowadays whenever I see a port or something interesting, the first thing I do is try to determine if any AI was used. This involves either looking at the gh contributors, or reading through the code to see if I can spot any 'AI-isms' I recognize.

I can't do that for every single project, so I tend to only bother if it's something that stands out. The result is that I just default assume everything homebrew/port/reverse engineering related is slop (which I recognize is unfair to a lot of people). In a similar vein, it's why I don't take Rust project announcements seriously anymore, since 99% of the time they're slop. Something about these two communities is apparently really appealing to vibe coders.

I don't mind if the site covers AI stuff, since most people here don't know/care what that means, and if something is interesting then I'd still like to hear about it.

EDIT: to add to the debate in this thread, here's a tangible example of one of the problems from vibe coded projects. Idk what motivates people to burn money on tokens to release projects like that, but that motivation won't last very long unless you're loaded.
 
Last edited by diggeloid,
It kind of is to be fair. For my curiosity I figured I'd try just straight vibe coding just to see what it could do before we got access to the various tools at work, and I have a seemingly functional PC port of Pokemon TCG with a custom duel overlay:

View attachment 587993

Doing this myself I do understand the need for like, better processes though. Handing Claude a repo of a disassembled game and saying "make a PC port" without having any kind of plan, any understanding of how it's done, or any idea of how to test it, will ultimately lead to a bad port with problems that you don't even know exist.


On this point, I just think transparency is good. People keep posting that they wish we mentioned a project used AI, and if there's a clear message on the repo, there's no reason why we shouldn't be including it in our coverage. Those people can then safely ignore the content, and people who are interested can engage with it. I know it never quite works out that way, but I think it is a positive step as a whole.
You caught my interest.
 
I have zero qualms about AI assisted code (I say that as a software engineer myself whose company also heavily pushes AI code). I'm convinced responsible use of AI code generation can make you like 25-50% faster (not 10x like some evangelists will spout, but meaningfully faster? yes). But you're responsible for the code it generates. Performance issues, security issues, that's on you. Fundamentally, code is logic and it just needs to work. I think that's a fair way to look at it. A good software engineer will design the architecture and have AI augment their own skill.

My main qualms with AI is generation of any user-facing assets. I generally consider that a marker of poor quality. AI remixes things it's already seen in training data. It's hard for me to look at AI pictures or music and consider that to be a meaningful original contribution on behalf of whoever is passing that as their work. I'd like to see an AI use disclaimer if anything user facing is AI generated.
 
Labelling sounds good. I have no issue using AI tools if that is all that is available, but I have no interest in using bad tools, human or otherwise.
Whatever legitimate use AI has, it has also revolutionised the production of low effort, buggy garbage. That is the actual main problem for me.

Like, for example, only the other day I wanted to use input redirection on my 3DS, but the only working Android client I could find seemed to be vibe coded. But it worked. They took an existing program, put it through an AI, and then either checked it worked or made the necessary fixes to get it working. Cool. Tedious grunt work seems to be the main use for it, AI assisted.

At the same time you have the Pokemon Gen 1 "Recomp", where someone took the existing code, shoved it through an AI, and then just dumped it on the internet for clout. Numerous obvious bugs, but who cares.
If that project was labelled as AI, it would have been a lot easier to tell the creator was untrustworthy. They knew that and actively went out of their way to hide it.

Granted that's an extreme example, its not on GBATemp to play AI detective. But labelling would be a start.
 
  • Like
Reactions: HaloEliteLegend
Labelling sounds good. I have no issue using AI tools if that is all that is available, but I have no interest in using bad tools, human or otherwise.
Whatever legitimate use AI has, it has also revolutionised the production of low effort, buggy garbage. That is the actual main problem for me.

Like, for example, only the other day I wanted to use input redirection on my 3DS, but the only working Android client I could find seemed to be vibe coded. But it worked. They took an existing program, put it through an AI, and then either checked it worked or made the necessary fixes to get it working. Cool. Tedious grunt work seems to be the main use for it, AI assisted.

At the same time you have the Pokemon Gen 1 "Recomp", where someone took the existing code, shoved it through an AI, and then just dumped it on the internet for clout. Numerous obvious bugs, but who cares.
If that project was labelled as AI, it would have been a lot easier to tell the creator was untrustworthy. They knew that and actively went out of their way to hide it.
That Android client has been around much longer than AI and always had those issues 🤣

Input redirection just sucks

Edit: I 100% misread
 
  • Love
Reactions: TheStonedModder
when a new recomp or whatever drops, wait for confirmation that it's human made.
dont cover ai, dont let it replace us.
I specifically disagree with this.

Proposing a news outlet delay publishing news is economic suicide. It means other outlets get there first and get all the clicks. Not to say this is all gbatemp is, but in the context of project coverage it's a lot of what we're talking about.

This isn't to say a news piece misattributing something - like saying an ai-authored item was made by a human - isn't awful, but leaving the ai-authorship of an item undefined during initial publishing is strictly superior to delaying.

You're not going to stop advances in AI by kneecapping the livelihoods of reporters who listen to you.
 

Site & Scene News

Popular threads in this forum