Good question - and that’s another problem. Not sure - can’t see any in Voyager. Can you see any on your end?

What does this mean? Are you saying we all have different tolerances and preferences for how much we value digital sovereignty?

Are you suggesting that we are all at the mercy of the Powers that Be and that they “turn the dial”?

Let’s use SearXNG as framing device.

SearXNG is a metacrawler. It’s not a search engine, it’s an aggregator. People talk about it like it’s a search engine but it’s not.

Which means your operation of it is beholden to others (quite a lot in that case). Sure, you control the hop. You don’t control what’s upstream of it. Whoever you use still gets pinged. Eg: Google still gets queried. Bing still gets queried.

You’ve added privacy at the point of entry, not independence at the point of source. What they do affects you directly, whether you want it to or not.

That’s something to think about if sovereignty is a concern.

Same issue with pi hole, email, VPN on VPS etc etc.

One thing I think would have helped would have been to indicate that your topic is AI-focused. For me, I went into the reading with the impression that you were talking about self-hosting in general.

It’s not (just) about AI. It’s the same pipeline. Frontier lab to local LLM. Upstream (Google) to downstream (SearXNG).

In both cases you’re running your own endpoint against infrastructure you didn’t build, can’t audit and can’t influence.

Self-hosting the bottom of the stack doesn’t change what’s at the top of it, and that’s exactly the problem…because the top directly influences the bottom.

The dependency just sits further back in the chain.

The chains bottom out at the same chokepoints - a handful of labs, a handful of infrastructure providers, a handful of index owners. No man is an island.

Anyway, I didn’t really want to pre-digest it this much, so I framed the post to allow out loud thinking in which ever direction it goes. This was not meant to be an exercise in “here’s what I think, fite me”, it was an invitation to think out loud together.

“vast majority don’t want AI anything” is doing a lot of work there. Do they not want AI autocomplete in the IDE? AI-assisted translation? AI-generated test cases? Because if the line is “any AI involvement,” the tag eats almost everything. If it’s something narrower, we’re back to: define the threshold.

This is a question about trust. That’s a hard problem because it assumes X is innocent, Y is guilty. I’m saying X and Y are both equally guilty (or innocent) until proven otherwise and Z (slop) can be summarily executed.

Treating a category of content as inherently suspect based on how it was made, rather than whether it’s any good. Having one tag [AI] puts a giant target on the post.

Thing is, the tag isn’t neutral metadata, it’s a flag.

And if 97% of projects have touched AI tooling in some form (who knows how deeply), you’re not tagging outliers anymore, you’re tagging the norm and implying everything untagged is the clean option.

That’s not curation, that’s a vibe-based blacklist.

I’m for “trust, but verify” - tagged or not. And the tagging won’t work because 1) what is AI coded any more 2) do you need the tag to spot slop (which is what I think people mean by [AI])

deleted by creator

Fair. So we’re talking about slop - and to that I agree.

Next question then becomes: does [AI] meta-tag in any way help you discern slop from non slop? There are comments and proposals in favour of it.

The counter argument is - slop is obvious…and even if it isn’t (and you’re going stick the thing on your own rig), you should probably do your due diligence first…which will uncover issues.

At which point, a scarlett letter isn’t going to do anything useful, and may unfairly tarnish projects.

That 97% stat was from 2 years ago. It’s surely higher now.

I don’t think [AI] tag works, for a number of reasons, as others have identified. But the topic keeps popping up here and there, so it’s worth mulling over.

The post was fodder to stimulate discussion, but to make it clearer -

Self-hosting gives you real but bounded sovereignty.

You control your stack, but not the foundations it runs on.

Even the most committed local-AI person is downstream of decisions made by a handful of people in boardrooms they’ll never enter.

“Fuck you, I won’t do what you told me” is worth defending, but it has a ceiling, and that ceiling is getting lower as AI consolidates upstream.

That holds for self hosting anything else.

The community sometimes talks about sovereignty like it’s binary, but it’s a dial, and we should be thinking harder about who’s turning it.

Finally, i am pro self hosted AI. I am also an AI critic.

Clearer?

Including immich, home assistant, syncthing, docker…

Right?

If you make a tag for AI, what’s the purpose of making everything people want to see explicitly say (not AI)?

Nominally, to avoid discrimination.

People that want to use that shit, need to learn to be upfront about and realistic about how the vast majority of people view AI produced… Well, ai anything.

Thing is, I don’t think people actually know how AI is used. At all. I think they know how AI is used to create slop.

And if they can already spot that, then why should things need to be tagged in the first place?

Do we want to see this?

• [AI] Jellyfin
• [AI] Home Assistant
• [AI] Immich

Because, going by the letter of the law, there’s a better than fair chance those projects have used AI - githubs own stats support that.

https://github.blog/news-insights/research/survey-ai-wave-grows/

https://survey.stackoverflow.co/2025/ai

97%. That’s Github’s number. In 2024.

To clarify…you in favour of discrimination and prejudice against ai users?

On it.

But your verification method is reading the code, checking git history, evaluating the work. That’s exactly the merit-based review I’m arguing for. You’ve just added a tag on top of it.

Are you of the belief that [AI] tag does good? I’m not seeing it but I remain open. Is it to make reporting easier? Is it implicit crowd control? I don’t get the utility.

To make clear - I don’t mind disclosure. In fact, I might even be for it. At the same time, GitHub’s own surveys show the majority of developers now use AI coding assistance regularly.

Which means the reasonable assumption for any actively maintained project in 2026 is “probably had significant AI help” - and that includes FOSS self hosted darlings.

• [AI] Jellyfin,
• [AI] Home Assistant
• [AI] Immich

Sits weird, but if were following the letter of the law, that’s where we might land.

Ah but it’s not 15-20 years ago. Low effort for a bot or llm doesn’t look like “yep”, “agree” etc. And it doesn’t necessarily have to look like slop either.

Creeping someone post history is problematic too.

If there were an easy solution to this, we’d know. At some stage soon, we’re going to need some sort of non PII cryotogenic proof of humanity.

That isn’t here yet, so I’m advocating that multiple signals are better than one.

What do you feel works best and why? You the one with skin in the game, self stated custodial not dictatorial role notwithstanding.

OK, so you’re not claiming to fix culture, just to give constructive engagement a cleaner lane. I can respect that.

Thing is, how are you going to verify [NOT AI] tagged posts? Because if it’s the porn argument (I know it when I see it), then you’re indirectly arguing for the same thing I am, with extra work for you.

Hell, if [NOT AI] can’t be reliably verified, your two-class system collapses to one class: the labeled ones. At that point it’s not an information system, it’s a stigma mechanism.

More so, disclosure tags don’t tell the reader whether the code is accurate, safe or good - shitty work is shitty either way. If they can tell it’s slop, you don’t need the tag. If you can’t, the tag isn’t helping.

I assume all software in 2026 has had AI assistance and evaluate it on its merits irrespective.

I’m not sure a scarlet letter does as much work as you’re hoping for.

Sorry, no answers, just thinking out loud with you.

I’m willing to try it as a system, so long as it doesn’t get fossilised.

What social capital is being built by waiting over an active participation requirement?

Display of patience and good will, as I said.

If you can’t wait 30 days AND engage in actual discussion BEFORE posting your amazing project, that’s a clear breach of the unspoken social contract.

Again - I’m saying these things are cumulative.

30 days + posts + 10%.

Any one or the other can be gamed.

Sign up for 30 days, do nothing, post ad = you technically waited 30 days.

Posts = “sounds great!”, “I love Syncthing!”, “Have you tried Jellyfin?” (or better LLM crafted variants) and THEN hit with ad = you technically engaged with the community.

10% = 9 fillers, 1 ad = you technically met the 10% rule.

How

See above

Low effort comments would just be considered spam in the first place, so how can it be gamed?

“Low effort” is doing some work there. How are you going to define it?

Its all human (me) reviewed, through a posted process that is transparent and simple to follow. Whats the difference?

That’s exactly my point. A+B+C = less decision pain for you, less gray area for posters. So long as both sides are clear on what A+B+C constitute.

The issue with rule 3 was a completely subjective interpretation of what was selfhosting as determined by that mod, which then made the community believe there were hit and run posts happening, and the posters believe this to be a hostile community (it was).

Correct. And it stifled discussion. We’re not disagreeing here.

Nothing is being automated, I review every report. The tools aren’t there imo to automate much of anything successfully.

Still not disagreeing.

I’m not saying no here, to be clear, I’m asking what this actually adds because I’m not seeing it.

Multiplicity of signals, transparency and clear expectations.

Maybe. So let me steelman what I think you’re saying.

It seems like you’re making a case for some kind of loop - AI tag means you can’t complain about AI tagged posts, and if you do, your post gets removed.

Have I represented your position fairly?

My counter point is - if the hostility already exists without a tag, that confirms the tag isn’t a cultural fix.

If complaints about AI use become off-topic and get removed, you haven’t improved the community’s relationship with AI content, you’ve just suppressed the feedback loop.

Is that really the right direction, given the reality of AI assistance in coding?

The hostility doesn’t go away. It goes underground, or people leave anyway.

Moderating away dissent isn’t community health, it’s just quieter dissent.

The community reacting negatively to AI content is the signal. A tag that makes that signal removable isn’t a solution, it’s a lid on a pot.

I’m not calling for some grand destigmatization of slop, I’m calling for nuance and education.

Which ultimately is beyond the scope of a mod…but you’re the one wearing the cape. What sort of house do you want to run?

Let’s at least not do something that propagates the knee jerk “fuckAI and fuck you” Lemmy is famous for.

It’s a hard problem, to be sure. Maybe your AI tag is a fair enough step for now.

We might be talking past each other here or actually agreeing.

But yes, this belongs in a separate [META] thread. It’s a discussion that needs to be had IMHO.

Yes, that’s what I’m saying. So an AI tag at best does nothing and at worst buries potential useful projects.

I’m confused by your confusion. I can try restating it another way if you can tell me what isn’t clicking.

Waiting is the point. People who are here to post and run will neither have the patience, nor the good will, to wait 30 full days before accumulating the social capital to blab about their project.

I’m not proposing “bot signs up, waits 30 days, then posts”. That’s a known escape velocity. 30 days by itself isn’t the solution because it can be gamed (esp without karma).

OTOH, everything proposed so far can also be gamed.

I’m merely suggesting a human mod reviewable method, that is also transparent and simple to follow for users.

If you can’t automate this, you either need hard rules (which will stifle convos - which was part of the issue with rule 3) OR you need a defensible rubric.

You can choose which, but there’s a constraints based reason why fora keep circling this sort of solution.