LLMs are great but they are brittle to minimal prompt perturbations (e.g., typos, indentation, ...). Q: How do we create truly multimodal foundation models? A: Do as we humans do: text as visual perception! Enter PIXAR, our work at #ACL2024NLP! arxiv.org/abs/2401.03321

#LazyWeb In the prompt sensitivity literature, what are the best papers that show how model performance varies with differently structured / phrase inputs?

Very insightful remarks by (((ل()(ل() 'yoav))))👾. Looking back in history, what is going on is not new and probably not very different from the empirical revolution in the 90s or the shift to DL in NLP in ~2015 (of which LLMs are just a natural continuation).

LAION-5B is important reference research dataset for reproducible language-vision foundation models studies. We release Re-LAION-5B as a transparent safety iteration on LAION-5B which fixes issues and allows broad research community to continue using open datasets as reference🧵

Fun new paper led by Ingo Ziegler and Abdullatif Köksal that shows how we can use retrieval augmentation to create high-quality supervised fine tuning data. All you need to do is write a few examples that demonstrate the task.

I’m looking for great recent/almost MSc graduates in NLP/Speech/Theoretical Deep Learning, who are interested in doing up to a 6 months/1 year position with me as research assistant. They would be hired by DTU Compute.

Can you imagine working in a company that not only supports you, but celebrates you? 😍 Feeling all kinds of gratitude for being able to work Hugging Face .

Our responsible AI team is hiring 3 research scientist interns this cycle (2 in Montreal, one in NYC). We're seeking enrolled PhD students who are excited to spend their summer figuring out how to ensure vision and/or language models work for everyone! metacareers.com/jobs/532549086…

Presented our paper on compute-efficient large-scale distributed model training using cost-efficient large-scale distributed poster printing Conference on Language Modeling yesterday! Happy to chat about compute efficient training & more today and tomorrow if you are at COLM!

How do multilingual models store and retrieve factual knowledge in different languages? And do these mechanisms vary across languages? 🗺️ In our newly released paper, we explore these questions! 📄 arxiv.org/abs/2410.14387 🤗 Negar Foroutan Desmond Elliott 🧵👇

Recent work led by Constanza Fierro and Negar Foroutan on understanding the mechanisms behind multilingual factual recall

I am looking for emergency Area Chairs and Reviewers for the Language Grounding Track for the February 2025 ARR cycle. If you volunteer, you will have no duties during the regular reviewing period. Emergency review: 18-26 March. Emergency area chair: 7-11 April.