Adversarial Triggers For LLMs Are 𝗡𝗢𝗧 𝗨𝗻𝗶𝘃𝗲𝗿𝘀𝗮𝗹!😲

It is believed that adversarial triggers that jailbreak a model transfer universally to other models. But we show triggers don't reliably transfer, especially to RLHF/DPO models.

Paper: arxiv.org/abs/2404.16020

Verifying LLM outputs is a time-consuming but often necessary endeavor, especially in high-stakes settings.

We propose SymGen, a new way to simplify LLM text verification in settings involving structured data!

📄arxiv.org/abs/2311.09188

I will be recruiting PhD students for a start in autumn 2024 via the ELLIS PhD call; application deadline: 15 November.
Successful candidates will join CopeNLU as well as Pioneer Centre for AI
More information about the PhD programme: ellis.eu/news/ellis-phd…
#NLProc #XAI

What is the best way to use language models for ranking tasks, such as 'rank the following countries by size: 🇺🇸,🇨🇳,🇨🇦,🇷🇺'?

We study pairwise, pointwise and listwise prompting as well as unsupervised Contrast-Consistent Ranking (CCR):

preprint arxiv.org/pdf/2309.06991…

⬇️ thread

If you're at #IC2S2 , join me and Yevgeniy Golovchenko by poster 30 today at 12:30 to discuss our ongoing work on gender bias in digital diplomacy!

If you're at #ACL2023NLP and interested in bias detection and historical documents, drop by our poster at 11 am EDT at the #RepL4NLP2023 workshop!

arxiv.org/abs/2305.12376

Around for #eacl2023 workshops?
- Arnav Arora will present on cultural values in LLMs C3NLP arxiv.org/abs/2203.13722; 5 May 12:00
- Isabelle Augenstein will be FEVERworkshop panel; 5 May 16:30
- Karolina Stanczak will present on bias detection @ SlavNLP arxiv.org/abs/2304.05783; 6 May 15:20

Attending #EACL2023 and interested in Cross-Cultural NLP?
Drop in to the #C3NLP workshop!
I'll be in Elafiti 3, 12pm presenting work with Lucie-Aimée Kaffee and Isabelle Augenstein on cross-cultural values embedded in PLMs.

Representing Rycolab at #eacl2023 with Butoi Alexandra Andreas Opedal Karolina Stanczak Clemente Pasti and Kevin Du!

We are super happy to share that our paper “Measuring Intersectional Biases in Historical Documents” got accepted to #ACL2023 NLP findings! Stay tuned 🎻

Work with Nadav Borenstein @TheaRolskov Natacha Klein Käfer @[email protected] Isabelle Augenstein

#NLProc #CopeNLU #ACL2023

I am excited to share that our paper “Multilingual Event Extraction from Historical Newspaper Adverts” was accepted to #ACL2023 NLP ! Details to follow soon.

With @[email protected] and Isabelle Augenstein

#ACL2023 #Toronto #CopeNLU

Greetings from #eacl2023 👋
Arnav Arora Karolina Stanczak Nadav Borenstein Amelie Wuehrl Isabelle Augenstein

Excited to share our SlavNLP workshop paper on measuring gender bias in West Slavic language models! Kudos to the amazing Sandra Martinková and Isabelle Augenstein for the collaboration! 🎉

Happy to chat about it in the poster session at
#EACL2023 !

#NLProc

Hi #NLProc , does anyone know of a good python library to evaluate sequence labeling tasks from character offsets? Ideally something that allows evaluating with varying modes of strictness to count partial span overlaps?

Massive congrats to Dustin Wright for successfully defending his PhD on “Machine Understanding of Scientific Text”. So proud of you ☺️👏
Thanks to Serge Belongie Andreas Vlachos Smaranda Muresan for serving on the committee & Daniel Hershcovich for moderating
#NLProc #factchecking

📣 Looking to start a PhD or postdoc in #ML #NLProc #XAI in autumn 2023? I still have several positions available:

- 1 PhD & 1 postdoc position on explainable fact checking (deadline: 1 March): jobportal.ku.dk/phd/?show=1582…, jobportal.ku.dk/videnskabelige… Pepa Atanasova CopeNLU

(1/2)