I don’t disagree, but this is a wildfire of interest right now and there’s a lot of people not recognizing this facet of how gpts operate. You have to really vocally recognize it’s weakness so it can be mitigated (hopefully).

No, they give you an answer that should sound correct enough to enable them to score a positive interaction.

Why do you think so many GPT answers seem plausible but don’t work? Because it has very very little actual logic

That’s why we really shouldn’t call them “AI” imo

Ouija boards made of databases don’t really think

I use planka. It’s compatible with my internal SSO and it’s exactly what I was after.

That’s definitely fair. I’ll admit its not the lightest on ram but it hasn’t gone crazy on me yet. With that said, I’m running if on a home lab k3s, and haven’t remotely flirted with running out of ram

I’m also using the hell out of it all the time, so I don’t mind it using what needs right now.

FWIW, with 3000 links, my container users 787 MB of ram. I don’t think that’s crazy considering what it’s doing during the archiving process

I’m not sure where security comes in for this, but that sounds like a reasonable build to me.

LinkWarden also saves various archival copies of sites (as well as it can with anti-robot measures anyways)

I love my LinkWarden instance. I haven’t found it lacking for my needs, despite over 3000 links thus far