Sven Slootweg (soft-deprecated) @joepie91@pixie.town

@NanoRaptor "Here Are All The Standard Medical Narratives And Expectations You Should Ignore Because They Will Not Apply"

(The chapter is an indeterminate number of pages)

@serapath It's odd because your traceroute doesn't look very different from mine, and it suggests that you *are* actually reaching the network that the server is on.

Can you DM me the IP address that you tried to access it from? So I can see if there's anything in my server-side logs relating to it.

@jollysea@chaos.social To clarify, for point 1, when dealing with an external drive, the problem is likely the internal SATA connector/cable and not the USB cable.

@jollysea@chaos.social Any of the following (not exhaustive) can be the cause:
1. Cable not plugged in quite right, usually also will show as intermittent I/O errors in system logs. Solve by replugging.
2. Intermediate controller does not support SMART. Sometimes a problem with some external HDD cases in particular. Solve by replacing with known-good case/controller.
3. Firmware of intermediate controller is junk and sometimes randomly fails to work for SMART. Same deal, replace with known-good case/controller.
4. Weird *drive* firmware configuration. Most common in refurbished/OEM drives, especially those sold as being for a specific model of rack server. Can sometimes be fixed with hdparm magic. Sometimes requires vendor-specific (usually Windows-only) tools.

@rune The big lifehack for me (back when this still mattered for work) was realizing that if you line-dry your clothes instead of tumble-drying them, nobody is going to complain that you didn't iron them

@navi FWIW this can be something as trivial as someone falsely hitting the fire alarm, those are often wired directly into the station evacuation system AFAIK

@existential1 There are a number of 'local models' (ie. not running on someone else's server), but they're most likely all still trained through labour exploitation and mass scraping of other people's work (though of course the vast majority of them don't tell you what they were trained on to begin with). The energy levels required are different between models, but still very high for all of them.

I'd say that there are no good options, only slightly less bad ones, and that this will likely remain true forever due to the fundamental technical requirements of LLMs as a technology (namely, training data at a scale that cannot credibly be obtained ethically).

@molenaar @ibestuur Op papier wel. In de praktijk zijn er helaas nogal wat zorgen over de huidige technische staat, die ook relevant zouden zijn voor een overheidsorganisatie (bijv. de 'state resets').

@existential1 In my last evaluation about a year ago, zero open-source LLMs that actually worked existed (although plenty existed that *claimed* to be open-source, but weren't)