wilo108

joined 2 years ago
[–] wilo108@lemmy.ml 2 points 1 week ago

Yeah, I don't disagree. I'm still on the 10.x branch (will need to upgrade to the 11.x branch in a few weeks, after the annual conference is done), so I didn't experience the UI fuck up, but it sounds ominous... sigh...

[–] wilo108@lemmy.ml 1 points 1 week ago

Everyone's using a client; the snikket invitation link makes it pretty easy -- I just print the kids a QR code (or text a link to their parents), and they've all figured it out easily enough. It's not perfect, don't get me wrong, but it's the best I was able to come up with and it's been successful.

[–] wilo108@lemmy.ml 1 points 1 week ago

I tested this thoroughly, and confirmed it's the encryption overhead, no question. It may not matter when you're talking to a server over the internet, but I'm on the same 2.5GBe LAN as my N100 box, and going over wireguard is noticeably slower than without.

[–] wilo108@lemmy.ml 1 points 1 week ago (2 children)

I run Mattermost successfully for a small-ish academic professional org, and tbh it's been simple, straightforward, and really pretty good. I live in constant fear of the ever-nearing and inevitable rug-pull, though -- they do seem to regard the open-source/self-hostable side as something they do on a minimum-effort basis only because they have to, so I would be hesitant to go down this road if I was starting again.

[–] wilo108@lemmy.ml 4 points 1 week ago (2 children)

I host snikket (which is built on top of prosody) for my family and (especially) my pre-teen children and their friends. Pretty happy with it, all things considered.

[–] wilo108@lemmy.ml 4 points 1 month ago (2 children)

Connecting to my N100-based box on my LAN is measurably (and noticeably) slower over tailscale than without. The encryption overhead is not nothing, and it can be meaningful depending on CPU hardware. (To be clear, not OP's problem, just commenting on "Tailscale on itself isn’t a measurable overhead.")

[–] wilo108@lemmy.ml 7 points 2 months ago* (last edited 2 months ago) (2 children)

Recently had cause to use https://github.com/oubiwann/metaphone (the doublemetaphone algorithm) for trying to match NER results from an audio transcript to known/expected entities and it worked really well. First pass: exact match; second pass: (double)metaphone; third pass: rapidfuzz.

edit:

Double Metaphone tries to account for myriad irregularities in English of Slavic, Germanic, Celtic, Greek, French, Italian, Spanish, Chinese, and other origin.

[–] wilo108@lemmy.ml 16 points 2 months ago (4 children)

"Stop clenching!" - "Add milk to the chauvinist." - when I've stopped laughing I'll give this a try, looks good, thanks :)

did you consider metaphone matching?

[–] wilo108@lemmy.ml 2 points 2 months ago (1 children)

wow, thanks -- off to check it out (after I've upgraded to 0.16.0 too) 🎉

[–] wilo108@lemmy.ml 1 points 3 months ago (4 children)

Any interest in implementing the "double bang" thing that SearxNG does? I find it super useful - I have degoog and SearxNG, and the "double bang" is the reason I still have SearxNG as the default.

[–] wilo108@lemmy.ml 3 points 3 months ago

Shows a failure if the DKIM selector isn't on an arbitrary list? What's going on there?

view more: next ›