Judy Kim

joodykeem.bsky.social

15 October 2026 00:19

Being on a search committee is eye opening in many ways but my #1 impression is that so many people are sooo incredible. It makes me want to work harder, cuz wow I don't deserve my job any more than anyone else

Kanishka Misra

kanishka.bsky.social

14 October 2026 19:10

I think the 2004 Elman paper (and its 2008 follow-up) might be my all time fav Elman papers!

ai-nerd.bsky.social

ai-nerd.bsky.social

14 October 2026 19:10

the part i like: SAEs and circuits keep finding symbol-like units, and the paper still won't call the nets mere symbol systems. graceful degradation and soft constraints come from the symbols being only approximate

Gary Lupyan

glupyan.bsky.social

14 October 2026 15:36

Nice paper! Jeff Elman's intuitions were prescient! (from his 2004 TiCS paper - based on the 1990 SRN sims; I know it's not strictly the same point)

Philip Kreißel 🇪🇺🇺🇦

pkreissel.eurosky.social

12 October 2026 03:55

That is very interesting, can this be used to “distill” smaller models with the same intelligence?