eleutherai.bsky.social

07 June 2025 11:28

Can you train a performant language model using only openly licensed text? We are thrilled to announce the Common Pile v0.1, an 8TB dataset of openly licensed and public domain text. We train 7B models for 1T and 2T tokens and match the performance similar models like LLaMA 1 & 2

Karl Rohe

karlrohe.bsky.social

07 June 2025 11:27

imagine if computers didn't have windows. just an entire screen of the terminal window. This is what computers used to be! And this is what LLMs are now.

Eugene Vinitsky 🍒

eugenevinitsky.bsky.social

07 June 2025 11:27

We can’t build housing, transit, or clean energy. If you’re going to criticize “abundance”, you should have a serious answer for how we overcome this

Eugene Vinitsky 🍒

eugenevinitsky.bsky.social

05 June 2025 20:37

One theory of the “success” of academic Twitter is that it occurred because of politics. It gave people an excuse to keep up with politics under the cover of keeping up with science.

Matthew Yglesias

mattyglesias.bsky.social

03 June 2025 05:28

The correct primary sequence is for all the states to vote on the same day, meaning that in a broad field nobody gets a majority of the delegates and then the candidates & their delegates bargain to select a good ticket.