Varvara Arzt

NLP researcher at TU Wien working on LM interpretability. Views are mine ☮️

[Khoury College of Computer Sciences @KhouryCollege]

Many AI users are familiar with sycophancy, where chatbots fudge the truth in favor of what they think the user wants to hear. But new research indicates that AI isn't just inclined to agree with you — it can start to talk like you, too.

Read more: https://t.co/MBEllwTxFg

[Khoury College of Computer Sciences @KhouryCollege]

When AI models regurgitate user prompts, the resulting answers can be wrong or outright nonsensical. Curious as to why, a group of researchers decided to put those models to the test.

To learn more about AI semantic leakage: https://t.co/fp5SMTSLBB

[Varvara Arzt @wienergespenst]

Check out Global PIQA for over 100 languages! Happy to have contributed to the project along with @TerraBlvns, @allanhanbury & Sarah Sulollari by putting Albanian on the map ✨

[Multilingual Representation Workshop @ EMNLP 2026 @mrl_workshop]

Introducing Global PIQA, a new multilingual benchmark for 100+ languages. This benchmark is the outcome of this year’s MRL shared task, in collaboration with 300+ researchers from 65 countries. This dataset evaluates physical commonsense reasoning in culturally relevant contexts.

[Varvara Arzt @wienergespenst]

Check out our paper for many more details: https://t.co/wFAxVq27H2
This work was done with my wonderful collaborators: @TerraBlvns @allanhanbury @GaborRecski and Michael Wiegand

[Varvara Arzt @wienergespenst]

Relation Extraction or Pattern Matching? How well do RE models generalise to OOD data? We find that higher in-distribution scores do not necessarily translate to better transferability.
Pre-print: https://t.co/wFAxVq27H2

[Varvara Arzt @wienergespenst]

#3: Structural issues in RE benchmarks, such as single-relation per sample constraints, reliance on external factual knowledge, and inconsistent negative class definitions, further impede model transferability. [5/5]

[Varvara Arzt @wienergespenst]

@svakulenk0 Congratulations! ✨ Are you back in Vienna?

[Varvara Arzt @wienergespenst]

@simi_97k @emnlpmeeting @gneubig Congratulations Simran :)

[Varvara Arzt @wienergespenst]

Thank you for great collaboration, @allanhanbury

[Varvara Arzt @wienergespenst]

Benchmarks drive progress, but transparency is key. In our work, we dive into the transparency of RE benchmarks and leaderboards.

Sadly, I can‘t attend the @GenBench workshop at #EMNLP2024 due to visa issues, but happy to discuss anytime

pre-print: https://t.co/sXdRy5Bhgj

[Varvara Arzt @wienergespenst]

@WendaXu2 @emnlpmeeting Congratulations, Wenda!

[Varvara Arzt @wienergespenst]

@simi_97k Congratulations! 🌟

[Varvara Arzt @wienergespenst]

@skoularidou @icmlconf Please feel free to contact me if you need anything in Vienna!

[Julia Neidhardt @j_neidhardt]

The 3rd ACM Europe DigHum Summer School takes place @tu_wien from Sept. 2-6. Take a look at the interesting program & the great speakers at https://t.co/LGhPh4lu8h Application deadline is July 14. #DigHum @DrCh0le @rburke2233 @JurgenPfeffer @StavrosMakris2 @socialhack @ErichPrem

[Marius Mosbach @mariusmosbach]

How valuable is interpretability and analysis work for NLP research? We (myself, @DippedRusk, @tombrownev, and @megamor2) investigate the impact of interpretability and analysis research on NLP in our new paper 👇

Paper: https://t.co/oMrYPaetu6 1/7

[Varvara Arzt @wienergespenst]

@WendaXu2 It is very important that you share your experience openly! I also hope that future generations will not have to experience things like that!