Some very cool recent work from Chengxu Zhuang on visually grounded language learning.

Surprisingly, standard text/image repr learning gives great image reprs but doesn't change behavior much on language tasks---how can we use image data for better learning of *language*?

Congratulations to Stefania Milan stefania milan (@[email protected]) and Don Weenink UvA Maatschappij & Gedrag, Tobias Blanke UvA Faculty of Science UvA Geesteswetenschappen, Max Nieuwdorp Amsterdam UMC, and Linda Amaral Zettler of NIOZ Linda Amaral Zettler IBED – all recipients of European Research Council (ERC) Advanced Grants 🎉
go.uva.nl/3Jcwy5M

👶 BabyLM Challenge is back!
Can you improve pretraining with a small data budget?

BabyLMs for better LLMs
& for understanding how humans learn from 100M words

New:
How vision affects learning
Bring your own data
Paper track

babylm.github.io
🧵

BabyLM is back for round two! This time we have: 1. Multimodal track 2. Bring your own data and 3. Paper track! See the CfP for all the juicy details!

What happy faces! 😊 It was a wonderful adventure organizing this workshop together with Raquel Fernández! Thanks again to Andre Martins Ivan Titov Iryna Gurevych and ELLIS for proposing and supporting it at all stages! Thanks again to all participants 🔥 #NLProc #AI #ELLIS

Folks, some Conference on Language Modeling stats, because looking at these really brightens the mood :)
We received a total of ⭐️1036⭐️ submissions (for the first ever COLM!!!!). What is even more exciting is the nice distribution of topics and keywords. Exciting times ahead! ❤️

Thanks Andre Ivan Titov Iryna Gurevych and ELLIS for proposing and supporting the organization of this workshop! It was a pleasure to make it happen together with Raquel Fernández and all participants 🔥 #NLProc #AI #MFO #ELLIS

Circuits are a hot topic in interpretability, but how do you find a circuit and guarantee it reflects how your model works?

We (Sandro Pezzelle, Yonatan Belinkov, and I) introduce a new circuit-finding method, EAP-IG, and show it finds more faithful circuits arxiv.org/abs/2403.17806 1/8

New work where language models learn to ask questions? So they can better understand user needs? With an amazing method name? Oh, yes!

I'm excited to release Prisma, a mechanistic interpretability library for multimodal models like CLIP and ViTs. Incubated at Blake Richards's lab & in collab with Neel Nanda.

Recent mech interp work has focused on language, but many techniques transfer. Behold, the dogit lens:

#EACL2024 was an absolute blast! 🔥 Our own contributions: a great talk (Alberto Testoni) and a must-see poster (@ecekt2) at the main conference + a timely tutorial (@michaelwhanna) and an #UnImplicit workshop on Thursday! ☀️ Well done! #NLProc #AI #language

_Unsure_ about what workshop to attend? Join us this Friday at the #UncertaiNLP workshop

Location: Bastion 2 room of the Corinthia

uncertainlp.github.io/program

Thrilled to share our work as a keynote at the UncertainNLP workshop!

I'll be covering the following work (ranging from semantic parsing to VQA and multi-LLM discussion/collaboration) under the umbrella of using uncertainty to rephrase/refine inputs & select outputs

Short 🧵

I'm attending eaclmeeting in Malta!🇲🇹 Drop by our poster starting from 4pm during poster session C! aclanthology.org/2024.eacl-long… #nlproc #EACL2024 Sandro Pezzelle Raquel Fernández