I wrote a blog post on the current trend of game decompilations using AI and where we’re going with this. It’s all very promising but oh boy are the implications scary.
Pretty sure “decompiling” has always been the easy step - open the binary with a hex editor and dump the contents as some source code. Making sense of that is where shit gets really difficult - making it work was usually even harder.
Can we please just skip the middle-man and start open sourcing everything instead of burning all the world’s resources to achieve the same goal?
Best I can do is ban software decompilation.
I know a few games that are sorely out of date by active developers and the devs are always like “I don’t have time” or “It’s not that important” or “It’s too hard.”
And I’m like, why don’t you release the code you prick, you don’t care, so why are you still lording over it.
So this could be awesome.
For most games, the answer is “publisher contracts”.
Lol sounds like some of our internal teams.
They don’t have time to support the things they built or to have people help them build it, but they refuse to make the repo internal so that others in the org can help them with it. Instead, they just act like bottlenecks.
I have been trying to recompile a Windows v3.1 game. Once it has finished cooking, I will try to contact the OG game’s author for another freeware release of their work. While quite raw, you can see the bones starting to knit together.
My hardware is an 7965WX, 128gb DDR5 5600, x1 RTX 4090, and x2 RTX 3060. While powerful for a gaming machine, this is modest for AI workloads. I hope to eventually replace the weaker cards with a pair of RTX 4500, as that would afford me greater quality and speed.
This project has taken three days so far, and I am expecting another four to reach parity with the original article. If I had the hardware I wanted, I expect it would have only taken two days for parity to be achieved.
What’s the motivation for even doing this if not to learn? If it’s just to have a playable version of the game, emulation covers that easily. If you want to extend it, you still kind of need to know how the original is designed to understand what types of changes fit within the structure of the game.
Quenzar’s Caverns is such a simple game, re-implementing it is a great project for someone wanting to learn how a graphical application is put together end-to-end. It would even be a good beginner reverse-engineering project if you want to reproduce it precisely, given it’s a 16-bit application.
At the end of the week, you might have a reproduction of the game that you can play. But you’ll have missed out on the opportunity to learn enough about program design to understand why throwing a week’s worth of compute at this task is totally unnecessary.
The goal is to establish a workflow of working with AI, the programs, the behaviors of the models, and so forth. In any case, I always sucked at math, so learning how to program just isn’t viable for me. My strong points lie elsewhere.
My goals are not yours. I want to use AI to translate videogames and other media, reimplement games with modern QOL, and eventually, to have media of my own imagination given form.
this post is the strangest humblebrag I’ve seen in a while
I am talking about my personal experience with the technology. The intention is to inform people what the technology can and can’t do, and what was required. I want other people to see what I am doing, and say “I can do it too!”
I want other people to adopt AI and to pursue the things that they think is cool. Be it creating manga, having tabletop game companions, or recommendations for baking recipes, I want everyone to have a better life from collaborating with AI.
Be it creating manga, having tabletop game companions, or recommendations for baking recipes, I want everyone to have a better life from collaborating with AI.
We used to did all the above together with other human beings. Sounds to me that you just want to replace them with machines. Well, good luck. You’ll never find true happiness down this path, in my opinion.
Not sure what there is worth bragging about
Don’t send a fucking AI-generated message to the original developer
Dude. I didn’t say anything, nor even imply, about an automated message. Also, I previously contacted the author and received permission to distribute the none-AI version of the game.
Seriously, folk like you suck - because you are hallucinating bullshit, and fling it at people.
Yeah, this is pretty awful
if this is scary to you please go outside
The scary implications being that any asshole with a Claude subscription can create something really cool in concept, but will more than likely end up abandoned in a few months when they inevitably give up (because this entire scheme is inherently short sighted) and all that’s left is a codebase so machine generated that likely no human would waste the monumental effort it would take to make it human readable- or logical to begin with.
“AI is a tool”- correct, it is a tool to scam people, and this certainly isn’t evidence to the contrary of that.
I’m not gonna sing the praises of something creating so much waste. Learn how to code, actually contribute to these things. Cry harder.
this is a pretty ignorant and assumptive take
This is very conflicting.
I daily a macbook, and so games aren’t that readily available. To be able to play something like the original Syndicate (1993, SNES and other platforms) or Shadow Hearts: Covenant (2004, PlayStation 2) on my mac would be awesome. I’m very much a patient gamer, so there are so many old titles I’d very, very happily play through again.
But there’s no way this would be restricted to old, dead platform games.
AAA games now cost millions or even, according to GTA 6 rumours, over a billion to make. If a studio publishes a game, and a week later there’s a free AI clone of it, then there is no more AAA games industry. Actually, that also follows for some of the awesome indie games I’ve loved and enjoyed. How can you make money selling copies of your game if there’s a free, or massively undercut, version of it available? Obviously most gamers would like to support creators, but in this economy? They’d probably rather eat and pirate games.
Pirate.
P-AI-rate?
Pir-AI-te?
Psst. Check out OpenEmu if you wanna play SNES games on your mac.
then there is no more AAA games industry
Oh no
Next do Hollywood and the music industry
It means one of two likely pivots if you ask me: the resurgence of highly locked down consoles with hardware protecting the data, or - and this is the most likely path - devs will just resort to making online only games, where lots of the content is streamed in while playing (I mean assets, not the display output - though don’t doubt the latter, if it ever becomes feasible enough I’d posit that devs would migrate to that as well).
Think a “live-events” driven game where every pre-determined period of time you download the assets for that period.
Fortnite-ification of gaming, basically. :(
Cracks have always existed? You could torrent GTA 5 on release day, or at least the single player part of it.
Sure, but with ‘decompiled/recompiled’ games apparently not breaching copyright, you’ll be able to get GTA 6ish without having to worry about torrent clients and magnets and seeds and all that super complex stuff that most people just don’t want to deal with.
If, a week after GTA 6 comes out at £70/£90, Big Stealing Cars 6 comes out at only £20 and is 90% as good… where’s the money going?
If they can clone games with AI so readily (actually it takes a lot of tokens but still less than it was to develop) then I’d think they could put that into developing their own.
Of course, like you said, if someone could just copy it and put it out there as soon as you release it then there is no motivation to make anything if you need to at least recoup your costs.
Why would they waste time and brain developing their own games when they can just piggyback on someone else’s creativity that’s had proven success in the market place?
It’s not LLM decompilation is LLM de-obfuscation
We have been able to de-compile pretty much anything for alomg while now, what’s been harder is making sense of what comes out of it.
The two being conflated c o n s t a n t l y pisses me off
this right here.
the adobe suite re-release is a much bigger deal than just pirated software. it’s the first time we commoners have been able to use the models they trained with our stolen content to steal something back, and it’s glorious.
Another exciting area is open firmware synths. specifically the $70 fm-1 (I think Pi based) is blowing up with synth engines/groove box apps. Its exciting new area for OSS. Its scary for closed synths.
Someone on reddit just vibe-coded a free Roland MC-707 app that turns the Groovebox into a real arranging/performing monster.
Are these AI decompiled games any good? I heard the Adobe ones looked good but we’re actually really slow and buggy
Most of them are really good. The Adobe ones aren’t actually decompilations, they’re just AI-generated apps that try to copy the same functionality which is why they’re so buggy and unfinished. The AI decomps/recomps should be functionally the same as the original games, usually with some edits to uncap the framerate and make it work with higher resolutions.
What is the benefit of AI Photoshop clone over, for example, Photopea?
Photopea runs on the web which makes it way slower than native apps. It chugs a lot when editing a 4k image for example. The other main benefit is that the clone is fully open source whereas Photopea is closed source.
It chugs on big projects, true. Is open source though.
Not scary at all for open source games! Or open source software. What, AI os going to decompile linux and 1000 AD?
Can it decompile windows and make it Linux
It’s called ReactOS :)
We have Linux at home.
Linux at home:
Compile this fucking [expletive] [slur]!
I want to see AI turn boring white male game protagonists into awesome POC ladies!
What’s is the token cost to decompile one of these games? I want to fix a game breaking ‘bug’ in an old game are we talking $10s or $100s to make that happen?
Given it’s a game from 2004, it should be relatively easy to decompile, so if you’re the kind of person happy to tinker - it could be a great project to self-host and try to get done yourself?
That way the only cost is time, and a little bit of electricity.
I only just looked up some of the basic requirements, as it’s piqued my interest in trying to decompile something myself over the Christmas break!
Will read up on everything. I got a 9B model running, anything else was dog slow or didn’t run, even on a 16GB card (AMD though)
Its very unlikely you will have any success decompiling things with such a small model, unfortunately this is a case where you need a cloud hosted model.
No idea about AI costs, but there’s many humans who would do it for free, if you let them know what the issue is.
That one specifically - the game is ruined by parking, you run out of space for car parks long before you really fill up the mountain.
The car parks have coach spaces, but the seem to be broken and never appear. Either fix coaches or remove parking completely.
Will look into it when I have some time next week.
the game is ruined by parking, you run out of space for car parks long before you really fill up the mountain.
sounds like a case for public transport! :)
They already decompiled indiegames demo and used that to build a clone of it and ship it, faster than the indie dev can finish their game. It’s definitely an awful situation for small indie dev because they not only have to compete with bigger budget, their peers, and the f2p market, now they have to compete with cloners
