Brandon Thomas, developer of ArtCraft, a suite of open-source creative tools meant to replicate Adobe applications for free, has now announced apps to replace Microsoft Office tools such as Excel, PowerPoint, and Word, respectively, named GridCraft, DeckCraft, and WordCraft, that are created with Claude. Thomas has shared his new clones to ArtCraft GitHub, making them free for all to download.
These are clean-room re-implementations.
Do you think LibreOffice includes a 99% identical Photoshop clone?
Clean room implementation would be taking the spec, API calls, etc. & writing the code that implements them from scratch. This is de-compiling existing code & converting it to another programming language. Same source routines in another language.
It’s like taking a book written in Chinese, translating it to English, and calling it your own.
Bullshit.
Which they did.
This is not a decomp. They gave the tool all of the available specs, and the expected behavior, and let it run. They had a suite of test PSD files but no Adobe software.
Please explain to these other commenters how and why that counts.
May I ask where this is documented? I tried searching, but there’s just too many results to parse, and/or my Google-fu is failing me.
I find it hard to believe near full feature parity could be achieved in under a month for multiple major apps even for an experienced dev using AI. It’s a bit hard to buy.
This project is weeks old. I dunno what detailed writeup you expect to exist. But the project’s Github page says: “Implemented from public specs and observed behaviour only. No proprietary code, shaders or assets.”
There’s no secret sauce in Photoshop. Image editors have been done. Every individual feature is a tractable problem. You can now buy your way out of those, in bulk.
Which brings us full circle to not original code, but I get that’s par for the course with AI. However, I can’t see completing all those apps - with full compatibility - in that time span with nothing but documented specs (especially nowadays when full documentation has long become an afterthought that never becomes reality far too often).
Sorry, not buying that there wasn’t decompiling involved. Dude needs to prove his “work” by releasing every detail of how he accomplished it so others can replicate it before I’ll eat my words. And he needs to do so now - so current tools, hardware, etc can be validated by Joe Public.
Argumentum ad nuh uh is somehow worse than all the sea lions saying your definition wouldn’t count even if you believed it.
I didn’t say anything of the sort. We all know AI code isn’t original, and for the purposes of this discussion, that’s fine.
My issue is the idea that he’s replicated virtually every last detail of several major closed-source apps simultaneously in under a month using documented specs alone without “cheating” by using their code in some manner. Plain and simple. Prove me wrong, and I’ll issue a mea culpa.
Yes you plainly did. “Sorry, not buying that” is just brushing aside that they say they did exactly what you asked for.
What is so unthinkable about producing compatibility with a file format that has a test suite? Or cloning any particular feature from a high-level description and a screenshot of its dialog?
C-clean room AI?
So like someone wrote trillions of lines of training data just for this project?
Ooorrrr did a fancy search engine hype them up that they’re a cool kid now and they wanted to believe it so bad they’re willing to claim ownership over anything
No such thing as clean room LLM output, unless you wrote everything in its training model.
It’s in fact by definition as dirty as possible.
Right, because clean room means you’ve never seen any code in your life.
People - spicy autocomplete was not going to derive C from first principles. Neural networks are example-based. The model still obviously never saw Photoshop’s source code, because that shit’s not on Github.
I’m with you on your first paragraph - unless we’re doing some pioneering work, we learn it from somewhere.
But the possibility of decompiling means you don’t have to have access to the source just so long as you can access the code is some form - including the generated executable.
The models were built upon the data, code, art, and texts stolen from a generation of internet users and artists, and from generations of writers.
Do you seriously believe that these companies which built themselves on theft would stop stealing once their first model was built?
If a single adobe programmer (or even employee, depending on how lax their internal security is) used a cloud-hosted LLM at work, you can be assured that whatever code and data that malware could access from their computer ended up being part of the next generation of that company’s model, and eventually of any other’s.
Even if they didn’t, if any windows 11 computer had access to the code you can be certain it scrapped it and sent it home to be given to openAI and any other “AI” company willing to pay.
Thieves will steal. No LLM model ever started clean, and they only get filthier and filthier with time.
That’s still not what clean room means.
Y’all will bend over inside-out to pretend these models need a terabyte of linear algebra just to copy-paste a text file they obviously did not have.