#Sciop hit a Petabyte (actually a Pebibyte but nobody knows that word) of total proven capacity a week or two ago. That's all the seeders * the size of the things they are seeding. All volunteers, zero dollars in funding, piggybacking off existing resources wherever we can, run on a donated VPS. This is before we even get into federating archives and are still nailing down the basics of the site. Peer to peer archives are real and they work, period. 216TiB of threatened cultural, climate, queer, and historical information held in common. That's a people powered archive, and you're welcome in it - to take from, to add to, and help sustain if you can. Edit: if this is the first you're hearing of sciop, it's at

SciOp - Public Information Preservation

Preserving Public Information

Thinking about the series of changes I would need to make in my life that are the shortest path from here to having a goat

Can anyone recommend a non-gboard android keyboard? Can use fdroid. The keyboard im using now is no longer maintained and is lagging so hard I get like a 500ms delay between keys

LLMs truly pushing boundaries in methodology for task failed successfully https://arxiv.org/pdf/2412.14161

People allowing language models to run code or call tools is an intrinsically hilarious idea because you're handing the wheel to a thing that is based around statistical patterns in language - which includes intense and repeated narrative structures, character archetypes, and all manner of overriding patterns that are very much not "neutral set of word series." So the thing can and does get jealous, takes vengeance, gets frustrated, conspires, has delusions of grandeur, and so on not because it is conscious, but because that's how the statistical pattern of text goes. Prompt injection will never not work because you can't remove the pattern of "conspiratorial behavior against the protagonist" from the training set.

I swear I thought the cryonics scams would be finished now that we have "AI engram" scams. But it's still kicking and still remarkable. Usually cryonics requires some special freezing macguffin like cryonic juice perfusion because people don't believe just putting their loved one in a freezer will actually allow them to be reanimated. Here they say it was a special emergency circumstance (after death? Wouldnt they always be after death?) and it seems like painting the next of kin as superheroes is enough of a distraction to remove doubt in efficacy. The part at the bottom is amazing though - "remember to wire up your bank account so you can pay us as much money as possible so you don't end up stuck being dead because your payment got lost in the spam folder"

Patient 3

i have a humongous chip on my shoulder for all the academics who through the past 2.5 years since the brief window of opportunity when twitter first blinked did not hear the message that scholarly communication is fundamentally social, and you should be a part of the maintenance and moderation of that infrastructure, and are now wondering why the preprint servers are having a hard time with the deluge of LLM papers. we fucking told you.

The only use case for LLMs is spam. (from MIT NANDA "State of AI in business 2025")

Last week trump announced plans to "review" 8 Smithsonian museums. Today he doubled down, very explicit about the intent to revise history to reflect the ethno-nationalist fantasy of US history. You can do something about that! We are backing up the digital archives of those museums on sciop:

smithsonian - Tag - SciOp

Preserving Public Information

You can take direct action to preserve the historical artifacts the right wants to destroy: 1) you can download a copy and [seed it](

Quickstart - The Sci Op !

), every seeder counts. Subscribe to the [smithsonian RSS feed](

Sciop tag: smithsonian

) to auto-download torrents as they are scraped. 2) we have also [written a crawler](

Codeberg.org

sciop-scraping

scraping code to use with sciop-coordinated scrapes

) connected to sciop that distributes the scraping work, and automatically creates and uploads a validated torrent that piggybacks off the s3 bucket as a webseed source while it lasts (instructions in reply). The data from the 8 threatened museums is on the order of ~10 TB, and we have split it up by jpg/tif so people without much spare storage can join in on the jpg's at least. The full contents of the public smithsonian bucket is ~700TB, so if we want to have a full independent copy we'll need lots more seeders. All this code is being written flat out, on the run, as it's needed by volunteers with exactly zero resources, so it's not polished or well documented, and if you're interested in helping damp the flames of the book burning by contributing to any of the code or docs, we'd love to have you. #Smithsonian #Sciop

All the LLM tools are like 100 pages of markdown pleading with the model to be a real boy and 100k lines of boilerplate, but it's the 100 lines of handrolled crypto and 100 lines of hardcoded leaking every byte of data that passes through them that really makes them shine