As far as I know, in Europe you cannot claim a copyright on purely generated code. And the generated code I have seen so far is surely not a 1:1 copy of something to be found online (in contrast to things copied from Stack Overflow by humans). Did you ever see a LLM obviously copying copyright protected code?
Nope. Furthermore, claude-code strongly suggested I use a GPL3 license for the aDSA project since it relied on some of the GPL3 gnat code.
I never explicitly called you out.
Though I have to ask, why do you feel that way? I have noticed some AI users interpret criticism of AI as a personal attack.
Licensing issues notwithstanding, having useful libraries is more important to me than whether or not they’re AI generated. Whether or not you like it, it’s like trying to stop the tide to think otherwise.
I’ve been experimenting with entirely human written specification (.ads) files with entirely AI written .adb files. If I like my prototype well enough to rewrite the implementation from scratch, would it be considered human or not?
Even if a library is human written, if it’s completely unusable, there isn’t much of a point. This is why I emphasize people writing a useful executable and then extracting libraries and not just tossing libraries into the wind. Especially with the power of AI now, the smallness of the ecosystem opens a huge danger for a flood of tofu-dreg libraries and then people trying to build on top. We’re protected a bit from that with manual Alire pull requests, but I’m not sure how long that lasts.
Because of the 2nd sentence. You asked a question – which is OK and I would have still send you the list for you to make a decision if any of it is a „higher quality contribution“. It’s certainly is of higher quality then I could have archived without the use of A.I. The „Please check for spelling, grammar and improve wording“ prompt alone improves documentation (gnatdoc and the website) significantly.
But then added a conclusion before receiving an answer — you already assumed the answer will be no.
Does that happen? I have been helping out with pull requests for a while and and I have not noticed any libraries „tossed into the wind“. I also have not noticed much use of A.I. apart form my own code that is. — But then: How would I know who is using A.I.?
You see, if you are working on a pull request all you are seeing to the alire.toml file and output of the build pipeline. For new users we check the website to see if it is a legitimate project. But we don’t actually check the code itself.
I think you overestimate the power of the manual pull requests.
Guidelines for reference:
My preference would be that if it could be considered derived from the LLM output under US/EU/international copyright law then it be disclosed. That probably depends more on the content than the process, though, in the past, I have read that some teams reverse engineering copyrighted software made a point of using team members who hadn’t seen the implementation code to rewrite from spec to avoid doubt that might have opened them to lawsuits.
Since, in your scenario, you wrote the spec and the implementation, it sounds like it would not count as derived from LLM outputs. If you did not read the LLM-ouputted code, then it could not be derived from it.
Licensing issues notwithstanding, having useful libraries is more important to me than whether or not they’re AI generated.
Disclosing doesn’t prevent one from producing a library with LLM assistance. But also, humans have been producing useful libraries without LLMs for decades.
With regard to disclosure, the Software Freedom Conservancy’s LLM-use recommendations echo your concern about the speed of receiving useful contributions while categorically supporting disclosure. Similarly, rsyslog has an “AI-First” (pro-LLM) policy that also “values transparency” and requires disclosure.
Whether or not you like it, it’s like trying to stop the tide to think otherwise.
I like AI and LLM technology, and I’m astonished by its capabilities. My personal concerns are about the ethics of the currently prominent models and of the companies producing them, in three areas:
-
appropriation of copyrighted material without the consent of the copyright holders or against their stated terms;
-
ecological sustainability and, especially, global climate change; and
-
support by companies producing the LLM models for their use in warfare and the use of those products and other software they produce in current conflicts in spite of documented human rights concerns.
Recently, a Google executive resigned citing concerns similar to (2) and (3). (He pointed to Google signing deals with the Pentagon and abandoning its commitments to carbon neutrality.)
I would have liked to have found that Anthropic and its models were different. As we know, they are not different on (1), having just settled a lawsuit over pirated books. As to (2), they use Microsoft cloud and will soon start using Google cloud. I suppose (but am not sure) that their training is not and will not soon be carbon neutral. And as to (3), they partner with a company, Palantir, that has been implicated in possible human rights violations, and their commitment to human rights may not be sufficient. In their disagreement with the Pentagon, among uses in warfare, they only objected to their products’ use in fully autonomous weaponry. Their stated reason is that it is not reliable enough, which suggests that there could come a time when that objection could be removed.
With regard to copyright concerns, I have indicated that as an ethical concern, rather than a legal one. I won’t elaborate on that now. On the legal front, besides the possibility of new laws in the US/EU/elsewhere, there are still many open cases in the U.S., and I don’t know about the EU and elsewhere. It may be years before the legal issues are settled. In the meantime, there have been reasonable concerns raised about what that might mean for FOSS software that incorporates LLM outputs.
I previously mentioned that some projects have banned or restricted contributions using LLM outputs. I’ve read many of the conversations. Mostly they have focused on legal issues and technical issues (e.g., workload of the code reviewers and goodness of the code). It might be that we, technologists, are less comfortable discussing messy, subjective concerns, but I have been dismayed by the seeming widespread lack of concern about the ethical problems of LLMs evident in such discussions.
I don’t wish to hold back a tide, but to extend the metaphor, a tide can turn. Ethical LLMs could be developed, and there is one move in that direction that I’m aware of with the Kl3m model. (There is also Common Corpus, but it is only a dataset, not a model.) And I am eager to embrace ethical LLMs when they become available. But for now, I plan to avoid code that is not clearly free of LLM outputs. If it is not disclosed, that will only mean I will have to contact the maintainers individually, and use old versions of code in the meantime, potentially leading to more effort on both sides.
True, but at least I’m a lot faster with Grok at in the passenger seat. A lot of what I did in the last year would not exist without A.I.
Ada is used to build weapons. I once applied for a Job building air defence cannons controlled by Ada. Ada was designed under contract to the United States Department of Defense. If you create an Ada library it might be used in a weapon.
And it was mandated to be used across all government departments, leveraging re-usable code/components. Then they copped out due to C programmer backlash (“It’s too hard … too complex … I have to type (literally and figuratively) too much”).
Imagine the ecosystem we would have now if the mandate held. For one thing, OS’s would have been written in Ada, not C. Think of the multi-trillions savings in s/w costs.
Ada is a language not a product for sale. Some people believe that sufficiently advanced AI should not be used for warfare. Taking that perspective is similar to believing that nuclear fission or fusion, or chemical or biological agents should not be used for weapons, and it doesn’t require believing that it should be impossible to wage war. Furthermore, some people believe that the conduct of wars, other military operations, and civilian law enforcement should stay within parameters such as the laws of war and humanitarian and human rights law.
I recognize that not everyone will share my views on war and use of force. That is just one ethical concern.