Slightly relevant: Math plagiarism incident, or an incident a longer while ago: Flow chart plagiarism incident.
meanwhile, the Community Standards were updated, adding a section about collective maintenance, human authoring and LLM. you can find the “official” position on LLM there.
EDIT(addition): The section about collective maintenance, human authoring and LLMs is a guideline, not a policy, on purpose. Policies invite rule lawyering and detection games. The “authored” vs. “written” question below is an example. The answer stays what it was: we only merge and ship code in Delta Chat that is human authored and humanly understood, and we do not skip human code reviews. You may have asked LLMs for initial analysis, e.g. to find bugs. Beyond that, a human has the last word, also on interpreting the guideline. The focus is on caring and collective maintenance, on purpose.
We might revisit the topic next year, but for now, contributors are juggling dozens of pressing topics daily, relevant to real users in many places, and we want and need to focus on those. Most of us deeply care to produce high-quality software, and we know of many problems and issues that require care and attention. If you see something concretely going wrong in PRs, please comment there. It helps more than abstract arguments here. Thanks!
Thanks for the update !
I don’t understand these standards, feels like they contradict themselves. In the upper section it says that everything has to be written by hand (=no AI) and then a few lines below it says that the usage of AI has to be disclosed, so what is it?
to be very clear:
we only merge and ship code in Delta Chat that is human authored and fully understood. this has always been our practice, and it remains today. also, we do not skip human code reviews. contributions have gone through and will continue to go through human review.
still, you may have asked LLMs for initial analysis, e.g. to find bugs
Thank you for the clarification. For me personally that means that I will prioritize other projects. For me Open Source means freedom and that includes the freedom to freely choose the tools that I use because what should count is the end result and not the means to get there. Putting a general ban on a whole class of tools is the opposite of freedom. Combined with the fact that it feels like the project has no clear development path not being able to use any LLMs means that one risks investing weeks of time in developing features that may not even get merged.
I understand that the challenges that come with AI are huge for Open Source projects, but a general ban feels to me like rejecting reality and an indicator for an unhealthy project.
Regarding your statement
I have a hard time putting it into context with this message:
You make it sound like LLMs were never allowed to be used in this project, but this thread does not look like this is true.
To be clear, I strongly support that every project can make its own decisions regarding AI. I just want to highlight that banning AI is not a no-brainer as some seem to think. Banning AI will also have negative consequences for a project and in my opinion in the long term it will harm and potentially kill projects.
I find this whole discussion pointless anyway. What’s important is that every pull request is reviewed and the code is good. Whether AI was used in its creation doesn’t matter and is often impossible to tell.
Besides, AI is constantly improving and will probably soon write better code than most average programmers. This AI phobia leads to nowhere.
I disagree. What’s important is that we keep our freedom, and AI being a fascist project goes the opposite way. We want to be able to live on this planet, we want to have good relationships (not adversarial relationships) with others, we want to be able to live, to thrive, to share. AI hampers that. It creates even wider hierarchies than already exist by putting even more people in the “exploited” bucket.
Good code isn’t enough. Proprietary code can be good, yet we don’t care because it’s proprietary. Considering that it doesn’t matter how the code is licensed is the same as considering how it is produced: fuel for dominant entities to continue their domination. Deltachat and co would never accept proprietary code, however good it is, so it makes sense to me that it also doesn’t accept AI however it’s used.
I know the policy isn’t there (yet !). It’s somewhere in the middle and I acknowledge it, at least it’s clear and we can decide where we’re going starting from that.
If your answer to a community’s documented expectations is “you won’t be able to detect my vibe-coded contribution anyway, he he he”, I definitely don’t want you in that community. Learn about consent.
I’ve watched this discussion as a non-coder who has embraced vibe-coding as a way of getting things done that otherwise would never happen. While I apreciate the back and forth, and can see the discussion from both points of view, the statement above is disturbing to me.
Many different worldviews are represented in this community, and I might be a minority here, but frankly, calling AI “fascist” is simply nonsense. AI, like any other tool, including a private chat/mail messenger, is neither inherently good nor bad. What matters is how it is designed, controlled, and used.
Up to now, the discussion has largely been about how AI is being used, and I think I appreciate the human verification and oversight path being proposed. That is a legitimate discussion. But assigning an ideological or moral character such as “fascist” to the technology itself is neither helpful nor accurate, and it unnecessarily turns a technical discussion into an ideological one.
There are plenty of legitimate concerns to discuss regarding AI without describing the tool itself in political terms that many of us simply do not share.
Unfortunately what has been missing for so long in tech communities is the understanding that tech isn’t an abstract world completely cut from the rest. We really need to learn a thing or two about sociology if we want to say things that make actual sense and are not just based on our made-up vision of the world.
There is no such thing as “just a tool”. This essay explains in better words why: There's no such thing as Just a Tool | deadSimpleTech
Every tool is the product of its creators, who are the products of their environment. Their usefulness stands on their inherent functions but also on thevery existence of the tasks they solve; the task must exist for the tool to be useful, and this is where it gets problematic.
If you want to talk about the tool you need to talk about the llm. Yes, the tool does things a tool does, you need to make sure it does what you want, etc… But that tool doesn’t exist without AI, ie the political project that’s only aim is controlling and managing the world, and in its current iteration has literally said democracy is not something we should keep, that billionaires shouldn’t be able to do whatever they want and everyone else should work for them. You cannot have llms without this political project, because it is this project that puts data in so it needs to amass as much data as possible; this project exploits the global south at a rate that would be illegal in western countries. This projects destroys the environment, requires ridiculous amounts of energy and breaks communities down to build datacenters. It removes agency from its users and puts it all back in the hands of its makers. Local llms are no different, they are not possible without the massive exploitation above.
You could say “it’s just capitalism/racism”. Yes, good, we agree that it’s bad. So, since llms are new and making things worse, maybe we shouldn’t use them ?
You may not share this political view, and that would not be surprising: fascism has been put in place not by people voting for it, but by capitalists specifically calling them into power and by enlightened centrists who would try to find a compromise between fascism and a more equal world. Unsurprisingly, that compromise is never enough for fascism, who always demands more. And centrists fall for this, again and again and again, until fascism doesn’t need to pretend.
It is up to anyone who likes democracy, however imperfect it is, to actively recognize the fascism playbook and combat it. Whenever it is there, openly telling you “I’m fascist and I’m coming”, to say no. Not another compromise, an apolitical cop out or something like that. We need to keep our freedom and for that we need to have a stance that makes sense. It is not enough to decouple llms from AI and pretend llms are fine actually. LLMs will give legitimacy to AI which will give legitimacy to fascists. The best step to take is to refuse it from the beginning.
EDIT: here’s a real-world account of how letting AI in, with what seemed like a good neutral stance, actually changed the community so much it effectively turned into a llm showcase.
I can accept the perfectly reasonable point that technologies are created by people, within economic and social systems, and that those things influence how technologies are developed and deployed. That is quite different, however, from concluding that LLMs themselves are inseparable from a fascist political project.
I think we’ve reached the point where our underlying worldviews are simply too different for this discussion to be productive. I don’t accept your premise that using LLMs legitimizes fascism, and I’m content to leave it there, as long as my view has been heard and understood from my own words, rather than through the ideological lens you are attempting to apply to me.
Only the future will tell us who was right
Writing “human-authored” as opposed to “human-written” here is a very disappointing, as is leaving out the code itself. Most people will read this guideline as condoning LLM use so long as the contributor is the commit author and the commit/PR message is hand written.
I think there should be community input. So many foss projects are condoning LLM use, and an inherently political project like Delta Chat should have an actual anti-LLM policy. You are interested in using Delta Chat in restrictive regions like Iran, so you should draw a hard line against using the infrastructure of companies like Microsoft and Anthropic that are actively killing people there.
LLMs are inherently fascist and capitalist because of their devaluation of labor, ecological destruction, etc, and the companies that build them are inarguably fascist, capitalist, genocide profiteers. It is impossible to fully avoid supporting oppression when using technology, but it should be minimized. Banning LLM usage is a clear choice.
In my opinion the whole discussion is missing the point. As I already said above, I am scared of AI and all the implications that it will have on humanity. I agree that AI is a huge danger for democracy because in the moment where human work does not have any value anymore the last bit of power will have been taken from the “little people”.
But AI is here to stay. It will not go away. Whoever tries to “solve” the problem by just banning it is not addressing the problem. Not even 1% of the users of an app will care if AI was used in the apps development. They want a great app and if they can not get it here, they will get it somewhere else. So is this project about making a great app that increases the freedom of people (i.e. because they don’t have to use Whatsapp and co. which btw. are adding AI to all their products and therefore users and their data are exposed to AI instead of being free with DeltaChat) or is this project about making a bubble in which developers can pretend for a while longer AI does not exist?
Software development is my passion since almost 20 years so it really hurts me to say, but whoever thinks that software development will still exist in 5-10 years in a similar way as it does right now is delusional.
Regarding AI what needs to happen is that we push for good regulations, for clean data centers, for universal income and so much more. Banning AI from a project because it’s “facist” (which it is not; even though most AI companys and figures are actually extremely questionable) will not solve anything. But calling for bans and making general statements is always easier than dealing with the hard problems that the future holds for us. Exactly these simple “solutions” are btw. something from facist playbooks, just look at people like Donald Trump or the AfD party here in Germany - simple answer to everything but not a single solution to any problem. Just ban immigration AI and all will be good 
“Fascism is when you reject something, and the more you reject something, the more fascist you are.”
I am really sad if just because something is more useful, we automatically accept it. We can argue the same with cloud and centralization, which DeltaChat supposedly be opposite of it.
DeltaChat is also not only about the app, but also about protocols used. In that sense, it is not just casual users actually using them, and protocols better discussed for actually developed and used by humans, not machines.
Contributors using AI will also require more money (if paying remote AI providers), or more powerful computers (if local AI models). The non-AI using contributors will be more limited and just give up.
The fashware (fascist software) movement already have their own DeltaChat competitors that are more AI liked, why DeltaChat need to join them?
The fascist also said they are for freedom, but at the end of the day, they just for doing freedom for the few (oligarchy), not for all people.
I think it is not pretending AI doesn’t exist, it is just finding AI negative enough to not be used in the main project.
It is always possible to have exceptions, like the non default clients can be developed using AI, but please, don’t make it a default.
“Whatsapp is here to stay, it will not go away”
“The cloud is here to stay”
“Proprietary software is here to stay”
“The metaverse is here to stay, adapt or get left behind”
This kind of statement is formulated as an abstract reality, almost as a threat, rather than on any analysis. There needs to be a proper analysis of the actual thing to know how solid the thing is, knowing that all of this is human-made and as such is human-removable. I’m pretty sure AI will collapse, because it is unsustainable, and capitalism will find another exploiting scheme to try to remain on top. I have all reasons to believe Whatsapp will survive AI, for instance. The cloud and proprietary software will unfortunately remain for a long time, because they have been core to the capitalist sociey for so long. AI is trying its hardest to also become a core pillar of all activities, before it crashes, so our best move is to fight it as hard as possible rather than already admitting defeat. All of this depends on what people do with it.
I don’t understand, what we’re talking here is a Libre software, using open standards, developed in the open. Everything here is counter to proprietary software, and yet there is no questioning that it’s not what we want to do. I really don’t understand this line of reasoning that says that just because something exists it only makes sense to use it and adapt. There’s no rationality behind it, and the only plausible explanation as to why this argument exists is that those using it want AI to happen and put it in the domain of inevitability. Which is why it is better to not take this argument as a really serious one.
That doesn’t mean DeltaChat has to use it. Seems irrelevant…?
If you talk about the actual services available, sadly seems like there might be a connection: David Culley: "After white supremacist Stefan Molyneux was banne…" - Hachyderm.io (Not that I’d know for sure.)
I have to admit, neither after reading this comment nor the policy I’m 100% sure what that means:
I assumed after reading the policy it means LLM auto complete as well as LLMs to directly alter or rewrite parts of the code are still allowed, as long as a human messed with it enough. The dependabot carveaout seems to match that. Is that what you mean?
(Not that I can change it or want to discuss it much. I hoped at least the rust chatmail core would manage a strict “no LLM output ever” policy to ensure the future of the project as much as possible. But it is what it is.)