Average Altaria Enjoyer

| Virgo | Any/All | 25 |

[sailermoon]

this is how new yorkers @ mamdani

[sailermoon]

[the-world-go-crazy2025]

either trump keeps staging assassination attempts to look tuff or they hiring the worst gunmen known to man JUST FUCKING KILL HIM ALREADY

[turnipstewdios]

Y'know, I kinda feel like when you get to the point that the general reaction to hearing about a failed assassination attempt is "Ah man, missed again. Maybe next time." you might be overdue for an impeachment.

[pigswithwings]

wtf are borders anyway. like yeah u were born on this beautiful earth buuuuut 😊 u cant go here. or here either

[walk-up-songs]

A forever banger from da share zone

[radicalurbanista]

you can’t support autistic people and support the police. No amount of paint jobs or sensitivity training can stop police from killing with impunity

[anarchopuppy]

And fuck the jigsaw puzzle imagery advanced by the eugenicists at Autism Speaks. We’re not a “puzzle” to be “solved“

Half of all people killed by cops have a disability, because cops aren’t trained to recognize or deal with people with disabilities and very few places have people who are trained for those situations easily accessible during emergencies. So the cops come in and do what they’re paid to do - murder anyone who’s inconvenient to the state

[rnismatch]

(backseating you at the mortar and pestle) man you aint even squarshing it

[imkeepinit]

[photoboybg69]

Exactly

[dogpuppy]

[evilsanlang]

it's okay if you as a lesbian want to fuck the straight blonde popstar but you can't be pretending she's a lesbian too girl at least make it a lesbian corruption kink or some shit

[evilsanlang]

"taylor swift is a lesbian" check your house for mold

"I can fuck the heterosexuality out of sabrina carpenter" questionable taste, but much more respectable

[twentybrokenipodclassics]

[suppermariobroth]

[coalashslurry]

we're running out of time to fuck it we ball....

[dogpuppy]

[damazcuz]

Are you doing okay? We missed you at the devil's sacrament. He mentioned you by name. Everyone was looking around and cheering until we realized you weren't there. If you need to talk I'm always here. At the aforementioned devil's sacrament.

[teaboot]

Idk who has to hear this today but sometimes the devil doesn’t need an advocate. Sometimes the devil is already winning

[teaboot]

Saying this as a recovered devil’s advocate btw, that bitch didn’t need my help he just wanted the exposure

[kubleeka]

[ashes2caches]

one time a guy friend told me he was quitting league of legends and literally two weeks later she was on estrogen. these events are intrinsically connected in my mind.

[eightyonekilograms]

There is still time [to stop playing league of legends]

[griffonatrix]

[g5-iguazu-deactivated20240408]

Hard at work at the yaoi factory

[g5-iguazu-deactivated20240408]

They fucking called fujOSHA on us

[tpwrtrmnky]

I think using a worse algorithm that consumes exponentially more resources than the alternative is morally reprehensible actually, and that is a general statement extending outside of just LLMs.

I'm talking video games that eat up RAM because the execs didn't want to pay for optimization time. I'm talking search algorithms that are made worse on purpose. I'm talking OS updates that break old hardware to run background services nobody actually needs.

Shoving "AI" into everything is just an extreme case of a wider problem.

[tpwrtrmnky]

Conversely, that also means that for some applications, machine learning models are the right choice because they're a reasonable alternative in resource usage terms.

[tpwrtrmnky]

While I'm at it, some data centers' cooling systems are actually fine because they're built in cold climates where the waste heat is used to keep houses warm.

The problem is the choice to build giant Heat Generating Machines in the middle of a fucking desert, and then using water that could go to watering crops for cooling.

And while lots of industries use data centers, the extreme need for parallel compute capacity over storage and transmission is a characteristic of machine learning that's troublesome, because the other uses of this extreme amount of compute capacity are niche and not immediately profitable. Lots of scientific work can make use of this type of data center, for example, but publicly funded research doesn't like overpaying for computer time.

The problem is essentially that tech is currently convinced they're going to make consumer services out of a type of hardware that's good for niche, specialized problem solving. They're trying to do this by selling access to a purported Ultimate Do-Everything Algorithm.

The incentives at play here are never going to result in a product that lives up to those expectations.

[tpwrtrmnky]

Anyway when I was working with supercomputers for my PhD one of my daily annoyances was waiting in the queue to run a large-scale physics simulation on like 4 GPUs, while literally half of the computer, several thousand GPUs, was occupied for days on end by a single AI training job.