Natanox

I don’t agree with your take on AI and the comparison at all, however if you want to use them and stay independent in the future it’s probably best to use Mistral. Their models are available for download and (mostly) licensed under Apache 2.0. You only have to pay for commercial use. Also they’re basically the only big EU-company in that space and the only I know of where the web interface isn’t infested with trackers and shit. However they are also involved in the military (guess which edge models are running on those semi-autonomous drones in Ukraine).

All I need to know is does it solve a problem I have, does it work, is it stable, and is it secure.

You have to be aware that

• LLMs are recreating licensed code without telling you, which WILL fuck you over eventually
• They do not produce secure code on their own. Keys end up client-side, in widely opened S3 buckets, encryption falsely implemented etc. Widely known, no link needed.
• It is not faster, in fact you’re slower while merely feeling faster. By now even the techbros themselves just recently finally admitted that.
• There’s no point to mention the immorality of the tech, everyone should know by now.

So yeah, your choice how much you use it. But it’s pretty obvious why nobody trusts vibe-coded stuff, and the metric ton of low-quality projects even forcing the de-facto App Store of a whole ecosystem to completely ban AI code reeeally doesn’t help.

I can respect the self-awareness though.

# Map any historical Claude co-author attribution to the project owner.

So I guess it’s slop? This looks shady.

Uugh, why do I see this at 3 in the morning. Good thing there’s Termux.

Security is the first thing that comes to mind. Compartmentalization prevents or at least makes it considerably harder for compromised services to screw up all the others.

Another thing would be that it might be easier to manage backups and snapshots.

One more thing I should’ve mentioned: It’s important to make a (free) account if you try it. Although they still offer a chat interface without it, that interface is then connected to one of their very small models and lacks the “Thinking” feature (same as ChatGPT’s “Reflection”). Not very useful in most cases and more of an appetizer.

Certainly interesting. I’d recommend you to take a look at Mistral AI (“Le Chat”), they’re a European company and far more trustworthy in terms of data security and privacy (GDPR and such) than US products. Their models are all available for self-hosting which might provide more flexibility in the future in terms of self-hosting and their web service doesn’t try to aggressively extract every data point from you (although you perhaps circumvent that with the proxy anyway). In my personal experience it’s also more likely than e.g. ChatGPT to admit when it doesn’t know something (or ask for specific data it needs) instead of making shit up, but I don’t have definitive data for that claim.

Of course I don’t know how well it works in Japanese or from Japan, if you try let me know! 🙂

Not quite

Totally fine. The only issue could come from legal implications since the domain registrations are managed by different organisations in different countries (leading to your registration data being an open book with .net domains but most likely unavailable with .nexus). However unless you’re silly enough to host a very gay social media instance using the TLD from god damn Afghanistan you’re probably fine (yes, that happened).

When vibecoders let the slopmachine set up a "My"SQL database and don’t check where it stores the API key.

Damn, the Lutris rule is spreading quickly.

I only know NextPush (Nextcloud App), but there is also something called Autopush I think?

Yeah, this is now inherently untrustworthy. Better to switch to an alternative.

Because it is for those who aren’t sysadmins or at least amateur Linux enthusiasts. The easiest tools quickly become very hard when something breaks and you got no one who could fix things for you you don’t know anything about.

They’re extremely new and open about what’s missing though. Their plans apparently got somewhat thrown all over the place by the sudden extreme interest and quite a few things aren’t yet in place (such as the self-hosting guide). Still works surprisingly well, and what they do goes into the right direction (no VC funding or investors, removal of the CLA, bound to GDPR, a full FOSS atack, etc).

Element is still as buggy as ever, unfortunately…

The only realistic alternative I’ve found so far is Fluxer, and that one is still in Beta. Very promising though.

Doesn’t your local DNS has to reach out to another DNS to query yet unknown addresses, causing an infinite loop as it’s now told to reach out to itself?

Might be a stupid question, but how are PR’s connected to your server deployment?

Renovate?

Good thing there isn’t a filter for “has working voice channels that aren’t a hot mess”, the list would be immediately and fully empty. With the exception of Mumble perhaps, but that one instead doesn’t have any text channels or community features.