I was never much into Nu Metal. Loved that 2nd Limp Bizkit album and Linkin Park's Hybrid Theory a short while during my early teens. Korn's Bass-slappy sounded cool but most of their songs didn't click with me. Talented as fuck vocalist though. Slipknot was mostly annoying and way too much Pop for me, though Jordison's drumming was always impeccable.
Early 2000s also exposed me to Godsmack, Drowning Pool but I barely remember any of it. I Stand Alone? Let the Bodies Hit the Floor? I think that's all that stuck with me.
But there's a singular Nu Metal song that pops into my mind like every other decade and then I listen to it until I get bored of it.
It's Ra's Do You Call My Name:
HERMETICVM
hermeticvm@circum.space
npub1rfw0...k3t8
Not a Virtual Machine. Hermetic Heretic Hermit. Raw Dark Pure.
Teuto-Gnostic Rebel Extravaganza on a supersonic journey.
CTO of a boring tech consulting company. Founder at not-so boring circumspace.
Looks like some psyop campaign is being used against Proton?
https://www.reddit.com/r/ProtonMail/s/kFXTrKKYqw
I'm so hyped but what this dude has been building.
Autumn is beautiful.


Physical work not involving computers can be great.
View quoted note →

Imagine someone paying you to STFU.
View quoted note →
GM


I usually listen to Gnostic Radio so I was kinda surprised to see Miguel in full Elvis regalia.


Meanwhile, the agent did not just pick up my infrastructure skill (which tells it to use pre-defined tasks where useful that also gates critical groups from being run (and all scripts also require manual approval to apply changes that might break things). It actively extends the taskfile I use to run my entire stuff, puts useful repeatable tasks into the correct group, tests all changes before a run and then re-uses each task instead of running scripts.
I'd expect this to make it much easier picking up parked experiments from branches instead of having to rela on agents appending a lengthy changelog or progress file which tend to become rather bloated.
Now this command lists all experiments run on this machine for each branch per benchmark run, with daemon being the name for the task group as well as my inference endpoint.
task daemon
daemon:task1
daemon:task2
daemon:etc
All benchmark results get their own section in the README.
Need to keep track of how this impacts token usage. Could go both ways. Expecting an initial increase in token use followed by a sharp reduction if I stick with the pattern.
Let's get Qwen3.8 Flash Next running on this thing. Not expecting performance worth using but it's a good exercise for my benchmarks and I get to learn more about this stuff. Also, I'll get to enjoy it so much more once that 7900 XTX lands.
I requested a reason for the ban, telling me which rule I broke. They declined. I requested an automated archive of my data from their systems which requires being able to sign in.
You’re forcing my hand, folks.
View quoted note →

99% of streamer's YouTube comments a nothing but bots right?
RIP Patrick Stewart


Asking a ladyboy on a dating/hookup app about their reasons for deceiving men with the “woman” gender label when there’s a “trans” option is a sure way to get your account suspended quickly.


Talked my mother out of getting some vaccines. Neither Flu shots nor COVID but I wouldn’t trust any needles since 2020.
GN

