Thank you for having me! I enjoyed being able to share my recent work and get valuable feedback from bright minds who have been doing a lot of inspiring work on understanding the social capabilities of LLMs :)

📖For our weekly MilaNLP lab seminar, it was a pleasure to have Minje Choi presenting 'Towards evaluating and Measuring the Social Capabilities of Large Language Models'.

#NLProc

#ThrowbackThursday #NLProc

'XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Models' by Paul Röttger et al. XSTest helps identify systematic failure modes in language models' safety balance.

arxiv.org/abs/2308.01263

For this week's MilaNLP reading group, Emanuele Moscato presented 'scGPT: Towards Building a Foundation Model for Single-Cell Multi-omics Using Generative AI' by Haotian Cui et al.

Paper: biorxiv.org/content/10.110…

#NLProc #ReadingGroup

📚 Happy to share that our work 'AnchorAL: Computationally Efficient Active Learning for Large and Imbalanced Datasets' has been accepted to NAACL 2024 (main) 🎉

Joint work with my supervisor Andreas Vlachos. Details in 🧵

1/9

There is a lot more detail in our paper, and in the dataset survey sheet. Go check them out!

arxiv.org/abs/2404.05399
docs.google.com/spreadsheets/d…

SafetyPrompts.com is now a paper!

We reviewed 102 open datasets for evaluating and improving LLM safety. We also reviewed how these datasets are used in major model release publications and in popular benchmarks.

Headline results and arxiv link below 👇

#MemoryMonday #NLProc

'The Past, Present and Better Future of Feedback Learning in LLMs for Subjective Human Preferences and Values' by Rose Kirk et al. surveys techniques for incorporating human feedback into LLMs and raises five unresolved challenges.

aclanthology.org/2023.emnlp-mai…

📖For our weekly MilaNLP lab seminar, it was a pleasure to have Xuhui Zhou presenting 'Towards Socially Aware and Interactional NLP Systems'.

#NLProc

#ThrowbackThursday #NLProc

'The State of Profanity Obfuscation in NLP Scientific Publications' by Nozza and Hovy tackles hate speech challenges in academic writing, proposing PROF, a multilingual tool for standardization to enhance accessibility.

aclanthology.org/2023.findings-…

For this week's MilaNLP reading group,
Derya Durmush presented 'KG-GPT: A General Framework for Reasoning on Knowledge Graphs Using Large Language Models' by Jiho Kim et al.

Paper: arxiv.org/pdf/2310.11220…

#NLProc #ReadingGroup

📖For our weekly MilaNLP lab seminar, it was a pleasure to have Fangru Lin presenting 'Graph-enhanced Large Language Models in Asynchronous Plan Reasoning'.

#NLProc

#ThrowbackThursday #NLProc

'Bridging Fairness and Environmental Sustainability in NLP' by Hessenthaler et al. highlights the need for research on the intersection between fairness and environmental impact in AI.

aclanthology.org/2022.emnlp-mai…

For this week's MilaNLP reading group, @Micky_Djiof presented 'Language model tokenizers introduce unfairness between languages' by Aleksandar Petrov et al.

Paper: arxiv.org/pdf/2305.15425…

#NLProc #ReadingGroup

Gender bias has been extensively explored in NLP applications like machine translation.

🤔But how can we use ✨emotion analysis✨ to investigate gender bias and stereotypes in LLMs?
With your help, let's first provide a bit of context!
🧵

Hoy a las 18.00 CET daré una ponencia en español en el Hackathon de #Somos600M organizada por SomosNLP 🤖

🗓️ Registrate aqui: eventbrite.com/e/registro-hac…

#MemoryMonday #NLProc

'Women's Syntactic Resilience and Men's Grammatical Luck' by Aparna Garimella et al. Gender annotations in texts show performance differences in taggers and parsers, highlighting the need to address gender bias in syntactic tasks.

aclanthology.org/P19-1339.pdf

For our weekly MilaNLP lab seminar, it was a pleasure to have Enrico Bertino in person, the Chief AI Officer and Co-Founder of indigo.ai, a chatbot start-up in Milan.

#NLProc

#ThrowbackThursday #NLProc

'Proceedings of the First Workshop on Cross-Cultural Considerations in NLP (C3NLP)' by Sunipa Dev et al. Discusses how cultural contexts impact NLP models and brings together researchers from various disciplines.

aclanthology.org/2023.c3nlp-1.0…

How many safety examples do #LLMs need?
What examples are most useful?
Why is it unethical to kill Python processes?🤯

Our new #ICLR2024 paper studies these + more! openreview.net/pdf?id=gT5hALc…
We analyze safey/utility tradeoff (100s safe demos suffice) and exaggerated safety.

Great…