Data toxicity can lead to harmful model outputs — and since most evaluations focus on English datasets, we’re underestimating multilingual toxicity in state-of-the-art LLMs. Our team partnered with researchers from CMU School of Computer Science and UVA to highlight this gap: bit.ly/PolygloToxicit…

Uncovering irregularities and non-smoothness in LLMs' in-context decision boundaries!! 📜 arxiv.org/abs/2406.11233

Caught the World Cup for India. #INDvSA

Ahh! This 💙 x.com/vk_venom18/sta…

Runway Very cool! Runway we will be happy to benchmark the physical commonsense of Gen-3 on VideoPhy. We have covered Gen-2, so it would be nice to study the improvements in v3 over v2. Is it possible to share the videos for the prompts in our dataset and we can test🥹☺️

We have released our DCLM models on huggingface! To our knowledge these are by far the best performing truly open-source models (open data, open weight models, open training code) 1/5

Comparing Bad Apples to Good Oranges: Aligning Large Language Models via Joint Preference Optimization 📜 arxiv.org/abs/2404.00530 w/ Ashima Suvarna🌻 Gantavya Bhatt Violet Peng Kai-Wei Chang Aditya Grover (2/3)

Curious about whether and how LLMs can help us analyze experiment results? Discover QRData at the #IJCAI2024 AI4Research workshop 🍊 🗿 🏝! Zirui Wu will present orally on August 5th at 15:00 and showcase the poster at 16:00. Stop by and chat with him!

Work from Google DeepMind shows synthetic data from smaller, weaker models >> synthetic data from larger, stronger models for LLM reasoning.

Way to go Shirley and team !!

The latest UCLA NLP group!!

Hard negative finetuning can actually HURT compositionality, because it teaches VLMs THAT caption perturbations change meaning, not WHEN they change meaning! 📢 A new benchmark+VLM at #ECCV2024 in The Hard Positive Truth arxiv.org/abs/2409.17958 Cheng-Yu Hsieh Ranjay Krishna uclanlp

Cross-lingual transfer can be as easy as swapping model layers between LLMs! 🔀 Our model merging method can compose math and language skills by swapping top&bottom layers from a SFT’d target language expert into a math expert without retraining arxiv.org/pdf/2410.01335 🧵: [1/3]

iykyk

✨Excited about this work from Yufei Tian ✈ COLM and team! LLMs tend to generate stories that are homogeneously positive and lack plot tension as compared to human written narratives that are more suspenseful, arousing and diverse. Checkout more insights in the 🧵👇