If you think I missed something I that I would find interesting, let me know :-). I'll include it.

But still, let's not focus on explicitly including syntax information...

Analysis von activation patterns in LLM do not match linguistic structure.

I don't think it needs "fighting". There has been a lot of interesting research on this conference, beyond "can LLMs do X". But I really appreciated your questions after the talks :-).

Methodological convergence ... #acl2025

... And a clear call for action at the end. Work together, serve people in reality. Do it now.

Rieser: Alignment is intentional, plural and deep.

Rieser: Systems overoptimized for safety are not helpful. They refuse to answer.

Rieser: #AI users now also include vulnerable groups, and that's why it needs to be a careful decision to whom and how to align.