For the EALM Workshop
"On Assessing the Political Biases of Multilingual Large Language Models" by @lernerp.bsky.social Laurène Cave, @haldaume3.bsky.social Léo Labat, Gaël Lejeune, Pierre-Antoine Lequeu, @bpiwowar.bsky.social Nazanin Shafiabadi and yvofr.bsky.social, collaborated with the STIH lab

Thanks, David! I'm going to go delete my post (which for transparency stated full cuts to these) to try to avoid spreading incorrect info!

no thank goodness!

there’s some coordination through AAUP around visas. see eg www.presidentsalliance.org/wp-content/u...

happy to see that USM (which covers UMD) signed

thanks. my main concern at this point are students and junior faculty. working to figure out how to ensure they’re protected and supported.

thanks i missed that

Hi Kate-- I'm really sorry to hear this. If NIH terminations are any guide, we expect NSF terminations to be chaotic, so we're trying to track them and organize the information. If don't mind, please have your team enter their grants here:

airtable.com/appGKlSVeXni...

And please spread the word.

Oh absolutely true - just didn't brand itself (incorrectly) as RLHF :)

Maybe worth adding: the transformer was industry playing its role in scaling up academic research.

Parallelizable attention + non-parallelizable recurrence, both from academic research, were SOTA but didn't scale. So Google did what industry does: make it scale (by trading rec. for a bag of tricks)

Thanks for having me, it was great to visit!

Agreed. They do add "valid/reliable" to a parenthetical later, which is along these lines. But there's also definitely stuff cut out.