Personally I would absolutely not want to ever feed my documents into an off-device model, but the point of self-hosted software is that it does what you tell it to and they absolutely should include letting you make bad decisions. decide what’s right for you.

FTFY (although I share your judgment). No disagreement about the rest.

Care to specify what you mean by “that”? It’s been a lot.

Long-time immich user here, similar problem. The issue (for now) is that only the owner of the media has write access. With partner sharing, all you can do is read the media (and add them to albums).

What I do for now with “trash media” (e.g. photos of receipts and other stuff that one snaps pics of for practical reasons and then forgets): I add my partner’s trash to an album named “please delete”. I periodically ask them to go through it and remove anything they want to keep, then have them select and delete everything in that album in batch.

Why this may not be practical for you, OP: you have two “trash producers”, so you’d have to keep a “trash” album for each of them (and sort trash media correctly). This is tedious work and error-prone: if you have only one pic in you “trash” album that your account does not own, selecting and deleting everything will fail, sending you on a hunt for the one misclassified picture in a batch of >200…

What immich may do about this in the future, as according to their roadmap: in the past, they were planning to add “more granular permissions controls”. Right now, they’ve renamed that to

User groups: Manage groups of users and share albums with a group

I hope this will include permissions management as well.

Wither way, the feature is likely to come with a greater overhaul of partner sharing. Partner sharing, as of now, is partly “broken” due to past design decisions: you can’t

• write to your partner’s files
• share facial recognition data: each one needs to classify people in their pictures individually (e.g. both need to tag “Uncle Fred” individually in their own accounts)
• share memories: only pictures owned by you will show up in memories, not your partner’s

And before anybody asks: no, there’s no ETA…

Soon

And now, everyone!

“IN THE LAND OF THE FREE
AND THE HOME OF THE BRAVE!”

That’s the thing: if constituents only get their info (or opinions) through chatbots…

…by axing or redirecting (read: embezzling) funds once spent on actual science?

True, although we must concede there is a difference between the simple machine learning already present in paperless and LLMs where, depending on quantization, CPU-based inference doesn’t get you very far.

Hook up the TikTok repos and you’ll get downgrades every 20 seconds.

I do understand and share people’s hatred of corporate-owned, centralised cloud AI.

I understand (though share to a lesser degree) people’s ethical concerns about how these models were trained (copyright has been broken for a while now, this just exacerbates it).

But this level of outrage about using local models on an opt-in basis strikes me as hysterical.

Feel free to change my mind.

Man, that is so outdated! /s

So… Here’s another resource that will become available again once the bubble pops?

I’ve seen Immich mentioned around here before, but I’ve honestly never set up even a real homelab or sever before, so that’s a bit daunting.

I haven’t been in this space for too long either, but immich is what brought me here. I tried it out locally - no deecated server, just spun up immich on my desktop machine to try it out. And it was just so good it made me want to have a server and learn the ropes.

I’d recommend you try it out this way too. We’ll see each other on the other side of that rabbit hole. ;)

I’m the violence.

they forgot what actual competition is

That’s an accurate diagnosis. U.S. big tech companies all follow the same playbook: use VC to buy off or price-dump to death all of your competitors. Become a monopoly. Then enshittify and extract the maximum amount of wealth from b2b and retail customers by increasing the cost of your services.

Any sane country would have drowned these companies in anti-trust cases and carved them up. But political leadership at the time was too high on neoliberal free market ideology and the fairytales these self-styled “tech geniuses” wove around themselves. Now big tech has effectively bought off the U.S. government.

They’re in for a nasty surprise when they find out that the U.S. is not the world, and the rest of the world can do pretty well without the U.S…

Good luck expanding that to the rest of the world which is just discovering that, while not perfect, China’s open weights models are

a) good enough for 95% or all companies out there
b) a lot cheaper, also because they’re
c) self-hostable, meaning: they’ll always be available, unlike U.S. cloud AI which you’ll be cut off from whenever Diaper Donnie feels it’s time for yet another bribe.

I second this. Firmware on these controllers is really hit and miss. My first controller (gen 5, iirc) worked flawlessly out of the gate. My second (gen 6 (?)) would co connect and then disconnect again. A firmware update (on Windows, grml) did the trick back then (and now I’m not touching that shit again).

It feels like nitpicking, but there’s a relevant difference here. If you can only access the weights (and not the training data and the learning algorithms used), it’s hard to get (say) a Chinese model to freely talk about Tiananmen Square Massacre. (Just tried it locally on Deepseek R1; it’s doable, but takes some prompt hacking.) The models may be freely available, but there’s bias and censorship baked into them.

It’s nice that these things are open source

Open weights != open source

Well, they do have that fruit basket, so…

Great, I’ll probably do just that this summer. Thanks for the info!