@darkuncle Hmm, I added the link in the toot with an edit. Might not have made it everywhere though. Anyway, here's where it's from:
https://www.printables.com/model/1149385-print-in-place-robot-boxbot-b-01/files
@darkuncle Hmm, I added the link in the toot with an edit. Might not have made it everywhere though. Anyway, here's where it's from:
https://www.printables.com/model/1149385-print-in-place-robot-boxbot-b-01/files
@darkuncle For filament, I've been kind of all over the place? But, I seem to go between whatever is dirt-cheapest on amazon for stuff I don't care about and Protopasta for stuff I do
@wavebeem Heh, yeah, we have a back door, so I wandered around the side of the house 😅
@SuperSelena64 what have you doonnneee?!
I don't know how legit this is, but the notion of Ukrainians selling engraved fragments of Russian tanks on the internet as tchotchkes is pretty metal
So, not only were Meta dirty book pirates, they were also filthy leeches and tried to seed the minimum ratio
Supposedly, Meta tried to conceal the seeding by not using Facebook servers while downloading the dataset to "avoid" the "risk" of anyone "tracing back the seeder/downloader" from Facebook servers, an internal message from Meta researcher Frank Zhang said, while describing the work as being in "stealth mode." Meta also allegedly modified settings "so that the smallest amount of seeding possible could occur," a Meta executive in charge of project management, Michael Clark, said in a deposition.
Part of me thinks this whole thing could have been a lot cooler if Meta had pirated the books, seeded the torrents, and then subsequently released the LLMs they trained as torrents since they've already claimed it's "open source"
Still pretty fucky but at least consistent, I guess?
There is also a part of me who daydreams about a scenario where an LLM can be a gift exchange.
Like, assume LLMs are useful. (I know, I know.) We know they require enormous amounts of exemplary material crafted by humans. We know they require enormous amounts of power & compute (modulo optimizations).
I think we're also learning that the LLM isn't the crown jewel - it's what you do with it.
So, copy the material, burn the compute, give the model back as a gift. No one got paid, but now we all have the model. Fewer folks have to burn compute to accomplish similar things.
Part of me thinks many fewer people would be pissed off by this. Still stomps all over consent, but it's at least less extractive?
Google kind of worked this way, once upon a time. They indexed your stuff on the web and derived value - but at least they sent folks back your way in the process. Not everyone was happy with this.
I dunno, I'm admittedly brain-addled by a career in tech such that I have to mindfully resist the impulse that when I see a pile of data I want to feed it to a computer and seed all the torrents.
It's weird feeling like both a dirty commie and a filthy techbro some days.
@zacwest Hmm, well then, my next e-reader will probably be a kobo and most books from the library or bookshop.org going forward
Computer UI hot takes:
semi-transparent video screens exist as a concept so we can see actors' faces as they stare at a graphic dramatically - mostly useless otherwise
voice interfaces exist as a concept so we can hear actors talk through a problem dramatically - mostly useless otherwise
Now, if you want to talk about voice interfaces as an assistive technology, then you've got me.
Otherwise, I will shout "everybody shut up, i have to talk to my computer about this, nobody else talk"
@jimniels That's okay, we'll train Large-Depth Hole Models to enable AIs to bury us
Just got really startled by what I thought was a person waving at me from just outside my second-story window.
No, it was a bird who'd just landed on a really thin branch, causing it and the bird to wave back and forth across my window with much wing flapping