I have had an influx of these game decomps on my Instagram recently and was wondering why so many were appearing out of nowhere. Shortly after that influx, the open-sourcing Adobe Suit thing happened and again its on my feed. Most of these projects use AI to make it easier to decomply these things and understand why it was used but I am a bit worried about the implications. As in, will these decomps actually be stable? Will they be maintained? Will companies start lobbying for AI laws harder now, in a way it benefits them and not us?
I personally stay away from AI due to the many issues AI brings but I am a fan of people fighting big companies but is this the way to do it?
It’s the good old copyright Mexican standoff. If it is ruled that decompilations are illegal, that shoots freedom in the foot, and even legal clean-room decomps can be taken out, but also ai models could be much more copyright restricted. It’s both a win and a loss for everyone, as tightening the copyright noose is exactly what companies want, so they have a lot more ammo to shoot at foss projects and competition with.
The other way is that this is considered fair use, as the ai is not verbatim recreating code, it’s recreating its own interpretation of the code, which is a flying spaghetti monster that only looks somewhat like the real thing only if you don’t open the hood. This, of course, opens the door to being able to use ai models to break down things legally, a win for decompilation, a win for security researchers, but also a big L as well, as now obscuring code behind closed source is not good enough.
In either case, hackers and do-bads will run their own models and decompile things faster to find flaws and break stuff. Making it illegal doesn’t stop determined badders from doing it. Pandora’s box is already open, there is no undoing this without society as a whole getting washed away.
Waste of energy. I am not anti-AI, I think it’s good in software development when used in a quite moderate amount and the developer understand every single word it spits out. When I use Claude sometimes, I can never help not rewriting what it spits out, it just doesn’t code the way I would. It’s especially often bad with bash scripts, and I am not a bash pro at all.
These recomps, and even these recent PS5 emulators, are hella vibecoded, so I am personally staying extremely far away from them. Also, this Adobe suite is supposedly barely usable, so there’s that.
For older games, where source code may not exist anymore and modding and preservation are severely hampered by this lack of source code, I can see the use of AI to build reverse engineered source code as a useful tool. For modern or brand new games, it seems like an irresponsible use of the tool. And there will be all kinds of legal issues surrounding the posting of the reverse engineered code online.
I can also see this as having some utility for checking if closed source applications are borrowing code from open source projects without proper attribution and/or following of the open source license. I would think that commercial projects would see this as an especially scary prospect as they begin to incorporate AI into their development pipelines. Could GPL’d or Copy Left source code work its way into the closed source project and end up opening the company up to liability? If so, this sort of reverse engineering may be the only way that is discovered. Which, may mean that this sort of analysis has a very useful purpose. Though again, posting the resultant code online seems kinda sketchy.
As another tangent, I’ve personally used AI reverse engineering for examining malware. I’ve run across samples in my work which I could deobfuscate and reverse engineer by hand. But that will take considerable time (hours at minimum, days on the outside). During a security investigation, time is often a luxury we don’t have. Being able to use an AI agent with access to REMNux and the REMNux MCP can cut hours to days off the initial analysis. Sure, there are always issues with hallucinations and the agent just failing at its analysis (I’ve already seen both). But, it can give a first pass at something very fast. That may mean IoCs and TTPs we can hunt on quickly to validate scope and containment.
AI is a tool, it’s going to be used for things both good and bad. And it’s going to take us a while to sort out the details and safeguards around it as we go.
There are a lot of mature open source projects out there. I wish people would put their time there instead of paying the slop machine to shit out an un-maintained copy
I’m of the mind that we are early in the timeline of AI that develops software. In 10 years, I believe that people will not do any software development. AI will be far more mature and you won’t even have the same constructs that we do today, maintaining projects won’t be a thing because it is only a process that exists out of current necessity.
We have grown accustomed to these temporal and defect triggers for making changes to software. That concept will eventually become obsolete. Instead of versions and builds, the development will just be a continuum, an ongoing process that doesn’t really end. Future software will not have these periods of time where nothing is done, it will just be a smooth stream of ongoing improvement. It’ll be the difference between watching a stop action movie that runs at 5 frames per second to one that goes at 60 frames per second. It won’t get updates on an irregular basis, it will just grow and evolve like ant colonies do.
There may still be humans involved at some level, maybe just to monitor and direct, but the need for “maintainers” will not exist as it does today.
I don’t, I believe that in 10 years society will be well on its way to collapse from severe climate events, food shortages, and war. Anyone who pushed AI will be dead along with the data-centers.
They can train AI on the entire internet and all it’s copyrighted content, that’s no problem they say. But we train AI on copyrighted code and it’s different? Yeah seems like something they’d cry foul about.
This isn’t about training, is it?
It was never about training.
Free all software. Death to DRM.
Recompilations are excellent.
The slop-based recompilations are worse than useless because:
-
They discourage humans from making recompilations based on the hard work of understanding the target software, since “There is already a recompilation.”
-
Slop recompilations “authors” typically don’t understand the target software well enough to implement the typical 10% of functionality the LLM fails to furnish.
-
The LLM generates such unreadable code that not even god could debug it. So it might as well be a binary itself.
Where are the decompiles of Photoshop in the last 30 years before LLM then?
deleted by creator
No, I am saying, where the fuck are all those decompilations of multi million LoC projects? They had like 30 years head start. They didn’t exist in the capacity that opus 5.5 enabled. I think that much is pretty obvious.
Curiously, all those folk that scream and shout about slop claim that the code has to be garbage. Even if that were the case, would the average user care?
-
My opinion? Decompiled software has always existed in a legal gray area, it just hasn’t been worth ruling on because of how challenging it’s been to decompile software. AI seems to have allowed sloppers to get in on the action, and given how many sloppers have zero concept of nuance, they’re being very loud about their efforts. In particular, releasing decompiled paid software seems like the easiest way to get companies on your ass. You’re probably breaking half a dozen established and existing laws, especially since some of these projects are selling their decompilation. I expect the legal system to step in and rule in favor of the companies.
To summarize, just like in every other field, AI sloppers are ruining it for everyone
Burning tokens to destabilise monopolies -
Not too happy about our environment burning but creating mischief rather than creating bad flyers has to be a win.
Decompliation projects are cool when they’re used to extend functionality beyond what could be done in the original game and romhacks. It’s awesome to see Super Mario 64 with uncapped framerate and a modern camera.
But a lot of these projects kinda seem like they’re not adding anything that can’t already be emulated. I suppose just having a decompilation is step one that can enable future enhancements, but it seems like recently there are far more of these being slopped out than there are modders interested in doing anything with them.
The fact that a lot of them are just slop now is kind of the problem.
There was a team decompiling Super Smash Bros Melee. It was slow going and they eventually switched to using AI to do it. You can tell where they switched if you look at the source code they published.
The old sections done by hand contain function names like “HSD_PadGetRawQueueCount()” containing variables like “queue_count” and “interrupt”. The new sections generated by AI contain functions names like “Fighter_Spaghetti_8006AD10()” with variables like “x221F_b3” and “x2224_b2”. This is slightly better than reading the raw assembly, but clearly nobody in the group understands what the new sections control or how they work if they can’t even give the functions and variables proper names.
So what’s the point? Why decompile Assembly nobody can understand if the result is archaic C code nobody can understand?
Aren’t that the sort of names today come out from Ghidra? It’s been a long time since I last reverse engineered something but I recall the process was to dump the assembly with labels based on addresses and then slowly start relabelling things once I figured out what each routine was doing.
I guess they are trying to go the extra step and generate C which committed l compiles down to the same machine code blocks.
I kinda hope it leads to another nail in the coffin for genAI. Let the lawyers fight the genAI companies and people involved and hope it only leads to harsh punishments and air tight restrictions for the genAI companies/gears¹. That’s the ideal future, but in reality it’ll most likely lead to reverse engineering and decompiling software to be made illegal if anyone pursues this legally.
As for the software, considering you cannot even trust generative AI giving the right information even if it’s giving you source links, I would only use it on the most air gapped system possible to avoid the probable massive amounts of bugs and vulnerabilities that are in it due to the gear¹ “working” on it most likely not knowing what literally any of the code does. Even then, I wouldn’t even download it because I don’t trust that they haven’t somehow fucked up and put up a highly infected download. Plus, I ain’t being a statistic as someone who downloaded that slop because it would show that more users are downloading that probable accidental malware.
¹ Aster Principle Def.: human(s) who willingly use(s) generative artificial intelligence ( genAI )
What I think is this will now be the reason why plenty of “AI first” companies will rethink their stance on AI.
Yes it is the way to do it. There isn’t another way or it would have been done already.
Yes they have been lobbying hard for ages to get AI regulated so only they can control how it’s used.
Stable? Maintained? Some will, some won’t.
The corporations need to learn they will own nothing and be happy.
I feel that its karmic retribution. These companies stole everything then started generating code and through corruption they were able to set the precedent that AI is not copyright infringement. All the companies embraced it with open arms.
Now to see it being used to open up and fosswash these proprietary products and its very funny, I’m waiting for the big companies reaction. I can’t wait for them to clutch their pearls.
I’d rather see open source software get the rescources they need to “catch up” to the monopolies.
The stuff ai is spitting out won’t be easy to maintain and any securrity vunerabilites will go unpatched. Attempting to maintain this software with ai will just result in making more dataceneters and more communities will be fucked over.
However, if adobe and microsoft sucesfully claim ai has stolen their code, than they might set the legal president that AI has to follow copyright laws. Which could help artists (wishfull thinking, I know).
