The software developer behind an unofficial PC version of the N64 platformer Banjo-Kazooie has released an impassioned rant about using generative AI tools for urgent PC ports of classic games.
The developer in question is DarioSamo, a member of the modding group that released Banjo: Recompiled earlier this year. He argues that the idea that generative AI enables faster porting is either outright “wrong” or just plain wrong, explaining that decompiling a game – that is, transforming the original machine code into human-readable code – in order to release it on modern platforms is supposed to be an exercise in understanding and preserving how that game was created.
Recently, the Youtube channel Video Game Esoterica posted a video about a Banjo Tooie “recompilation” project that uses generative AI. According to DarioSamo (via VGC), the new genAI recompilation is based on “tooling” designed for Banjo: Recompiled, but is not the work of the Recompiled team. He feels VGE didn’t make this clear in its video, calling it “irresponsible” and urging generative AI users to “keep launching these projects just to get their 10 minutes of fame in your videos.”
In a follow-up thread on Blueskypublished last night, DarioSamo detailed the impact of people using generative AI to remove PC versions of older games. He began by explaining what a “recompilation” is – a type of port that uses “static recompilation” to automatically convert one game’s code “into code that behaves the same way on another platform.” These recompilations sometimes use complete or incomplete decompilation projects “as sources of information to make changes necessary for a port to work properly on the target platform,” DarioSamo continued, “including improvements needed for high refresh rates, widescreen, quality of life and more.” »
Recently, people have been using generative AI to automate things, but DarioSamo claims in his thread that “all claims that AI speeds up this ‘tedious’ process are false when a recompilation is involved”, and that developers were already fully capable of “getting a recomp working in less than a day when these tools were released two years ago.”
“The trend we’ve been seeing for several months so far is that a new type of person has entered the scene and thinks that because AI is helping them get over the hurdle of launching the game, the job is basically done,” DarioSamo wrote. “And all they need is a credit card and a subscription to an AI service. What these people don’t know is that running the game is basically the first step of the journey and not the end.
“This is why I argue that when it comes to recounts, AI has not provided any sort of speedup,” he continued. “Doing a simple recompilation has always been easy and required very little time. It’s the polishing and producing a quality, durable project for the general public that makes the difference. This is the part that actually takes a lot more time.”
To illustrate this, DarioSamo highlighted Reconnaissance arrow DK64 – a port of Donkey Kong 64 which does not use generative AI. “(The Rekongpiled team) was racing against another project that used AI,” he wrote. “As the developers were fully familiar with DK64, they were able to get the game into a very good state, speeding it up even further by giving them information on how to use the framework to its full potential. Meanwhile, the AI efforts failed to make further progress because they lacked this knowledge.” (DarioSamo did not specify which genAI Donkey Kong port he was referring to.)
Generative AI can enable faster decompilation, DarioSamo concedes in his thread, but only if all you care about is ending up with code that compiles to something matching the original game’s ROM file. “However, the point of decomps has always been to document the code and understand how it works,” he wrote. “That’s why decompilation and porting efforts are typically done by different groups of people: one group works to understand and document the game code, while another group may use that documented codebase to help with their porting efforts.”
In contrast, large language models “will often force this process and result in a result such that, even if compiled into the correct ROM, it is not readable or understandable by a human or even by the LLM itself,” DarioSamo continued. This makes it more difficult to create a port based on the decompiled code or patches for the port once released.
Using these tools also means shortening the process of working on code in tandem with other creators and port stakeholders, DarioSamo wrote, “which means that new people entering the scene also profoundly ignore a lot of small but crucial decisions early in development, leading the tools to create an unrecoverable project that must be restarted from scratch.”
The rush for generative AI ports “can demotivate a developer who was in the process of creating a project that checks all the boxes of quality standards expected by the public,” DarioSamo said in conclusion. “It’s important to show your support for these developers, and allowing AI-generated projects to be promoted in our spaces would only discourage them further. It’s very important to assure them that there is a detail-oriented audience and that things are being done correctly.
“These are old, beloved games that have been available in many ways for decades, so there is no rush to be the first to provide a solution using these tools, and it is unlikely that it will be the only solution in the future either.”
What strikes me most is the missed opportunity to understand and preserve how a game was made, in a creative sector already plagued by a cyclical brain drain and loss of crucial documentation. This is deliberate, anti-social obliviousness in the name of a (theoretical) quick comeback, which I think is very much in line with the genAI philosophy as a whole.
The discussion around using generative AI for recompilation echoes similar anger in modding communities. Recently, the creator of a Total War: Warhammer 3 mod “deliberately corrupted” his creation in order to prevent people from using it for “AI-related purposes.” In August, Mark interviewed NexusMod users about what they consider an acceptable level of genAI use and how platforms should ensure transparency.
This article has been edited to correct a misreading of DarioSamo’s post about being able to get a “recompile in less than a day when these tools were released two years ago.”
