None of the ideas are patented (maybe dt UCS comes closest). Reimplementing FOSS is cheaper now than before, but it has always been a possibility.
Plus, corporations rarely cared about FOSS: who would sue them, who would have the time and money? How is it even possible to prove if they copied FOSS code, from compiled binaries?
1 Like
Am I mistaken. I actually thought the main companies depended upon FOSS to develop their products.
It’s one thing to use a tool. Another is to ship a product that relies on e.g. open source libraries that have permissive licensing terms, allowing use in proprietary products. Those are both legal (many companies still fear the FOSS licences, though).
But there are open-source components that require publishing your own product as FOSS, if you rely on them.
This domain that @epeeist linked, would allow companies to take a non-permissive FOSS component, extract the (non-patented) ideas and algorithms, and implement functionality identical, closed-source code, owned by the company, all that in a cheap, fast and automated way.
2 Likes
I understand how this is awful, exploitative and very dishonest.
But the same practices have been around since FOSS exists, albeit not in such a trim package.
Also, I don’t see why I would need to use their tool when I can do it myself comfortably from my desk.
They are fear-monglers, not targeting you, but legal departments of companies, by taking over any potential legal consequences.
2 Likes
I would assume an AI (maybe coupled with ghidra) would at somepoint work the other way around. No more reason to buy proprietary SW
Interestingly, they’re now debate about re-licensing a rewritten FOSS library, chardet.
There are several twists that make this case particularly hard to confidently resolve:
Dan has been immersed in chardet for over a decade, and has clearly been strongly influenced by the original codebase.
There is one example where Claude Code referenced parts of the codebase while it worked, as shown in the plan—it looked at metadata/charsets.py, a file that lists charsets and their properties expressed as a dictionary of dataclasses.
More complicated: Claude itself was very likely trained on chardet as part of its enormous quantity of training data—though we have no way of confirming this for sure. Can a model trained on a codebase produce a morally or legally defensible clean-room implementation?
As discussed in this issue from 2014 (where Dan first openly contemplated a license change) Mark Pilgrim’s original code was a manual port from C to Python of Mozilla’s MPL-licensed character detection library.
How significant is the fact that the new release of chardet used the same PyPI package name as the old one? Would a fresh release under a new name have been more defensible?
As I understand it, a clean room implementation means that the implementing parties have had absolutely no contact with the code they want to replace. That seems not to be the case here, making any claim of a “clean room” substitute suspect.
And I think any clean room implementation of open source would have the same issues attached to it…
1 Like
I wonder if anyone would have even commented if he simply started his own library with a different name, and then stepped down from maintaining the original one.
I totally see how FOSS maintainers burn out.
Completely bizarre experience reading this, as I attended college with Dan, though I doubt he knows who I am. He was a year or two ahead of me and I only ever briefly met him in the computer lab.
1 Like
Gemini is particularly bad (or good?) at laundering code. I write a lot of power shell, and the number of times Gemini has lifted code directly from a blog post, changed the variable names, and presented it is high. The other times it is making up libraries and methods.
3 Likes
That’s my boy 
1 Like
Gosh.
The blatant irony of ‘liberating’ open source software…
Yeah I don’t think so…
I was also impressed by the testimonials… like from “Profit First LLC”
1 Like
… that website is just obvious satire, isn’t it?
6 Likes
I am afraid it is not. Open Source, but specifically GPL code has been very frustrating for some commercial organisations. They consider the GPL license to be a virus. GPL is not free they reason, because you are forced to publish changes under GPL as wel… Why do you think that Apple is running more BSD like kernel (see the Darwin operating system) and not linux? A big reason is the license. Or the move from bash to zsh… is mostly because of licenses (GPL-3 is a problem for Apple)
If you look into the world of front-end development (React, Angular, etc) almost al licenses are MIT (or Apache). Basically, it is free to use to do with it what ever you like. Just don’t expect anything from us.
So, if you can create a LLM that is creating the same code - functionality wise - and ‘break the bound of GPL to be used in none-GPL applications’… Yes that is ‘liberation’. (NB. this is NOT my view)
But talking about this subjet of licensing… Is LLM generated code an issue for darktable license wise? Has anyone done researching in this? It can be argued that LLM output has no copyright, so is this a problem in a GPL3.0 context as darktable?
1 Like
Isn’t it true that code from LLMs cannot be copyrighted anyway?
Yes, but that’s not an issue if its some internal library for their app that they don’t want to distribute anyway.
Note that LLMs are also ideal for reverse-engineering binaries. Making up sources for any program is either feasible already, or will be soon.
I think @Masterpiga module is a perfect example of the worry: it seems to work, he can obviously code, but he has stepped into a domain that he admittedly doesn’t know much about.
How is he going to be responsible for this module when he doesn’t know about the domain in which it functions? Does this shift the burden to others who do know? Do those people have the bandwidrh? Who’s responsible for fixing bugs?
1 Like