Illustration done for the 1993 San Diego Comicon program book celebrating the 60th anniversary of the release of King Kong. #Godzilla #kingkong #Gamera #ymir #cyclops #cthulhu #swampthing #manthing #thepile #wolfman #nosferatu #happybirthday #monsters #art

The vendor pile begins 😂💗🌈

#colorsthatbuzz #strangelycuteart #artvending #thepile

April illustration from the 1994 Misspent Youths Swimsuit Calendar. #misspentyouths #thepile #punk #rexocola #comix

ghnggfhddhhhcckdfggnfdjddcj iii iii it ii i just i that idjsnssk thanks for sddingg to thepile

This felt like a lot of "gotchas", but I'd actually be very interested something like this for random bits of ThePile or FineWeb

Split wood
.
Put wood in the gap in the rick
.
So it's time to buck more logs
.
And repeat
.
#pottersofbluesky #ceramics #pottery #woodfire #anagama #noborigama

No invite. I got the best off-format Pile ready to go!

moxfield.com/decks/q_fMQT...

ya cayó el de #nodos, así que este va a #thepile

ya ves, con uno de hadas bordes que tengo ahí en #thepile y yo perdiendo el tiempo

💁‍♀️ Green cllrsl continue to gather info about options for the old bus station site & to press Labour to provide transparent information about plans as soon as possible.

💚 We have long called for this prime city centre site to be used creatively for pop up market trading, green spaces & play areas.

Vintage: Lurrus PO
Legacy: The Epic Storm, Forge
Duel Commander: Cloud
Pauper: Boomer MonoBlack
Extended: NoStick
Pioneer: Greasefang
Modern: Ral
Standard: R/G
PreModern: Rock, Tinker, Tog
Misc: ThePILE (type 4)
Cedh: Every deck
Edh: Go-Shintai, Neheb

Favorite Card: Yawgmoths Bargain

2/🧵 Our study systematically explores how demographic biases embedded in pre-training corpora like ThePile affect LLM outputs. We reveal substantial misalignments between LLM representations of disease prevalence and actual data across demographics. 📷 👥

You’re right these results apply to general “big” datasets like ThePile or RedPajama. There are several papers at ICML on weighting datasets like Chameleon (icml.cc/virtual/2025...) that could probably let you get away with less data.

The claim is that the LLM's training materials / dataset included copyrighted works (including those of the authors bringing the lawsuit).

Meta has publicly identified sources for books that were used in the training dataset, including Project Gutenberg and Books3/ThePile.

all is a big thing but you can find
- historical Common Crawls (2016-2026)
- torrents of the dataset(s) that were the inputs of llms released in 2021-2022 (eg ThePile)
- torrent anna's archive and filter on dates. I don't think there are historical datasets

Alderman... jo, me suena y no sé de qué. Pero tengo ganas de leer cosas nuevas. Pa #thepile que va

I wonder if it makes sense to re-pretrain gpt-oss, or long-posttrain it with aux ptx loss on domain-prefixed ThePile garbage. Might make for a strong model