🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Wonder why DPO works so well? Check out our paper led by Yunjae Won for deep insights into its effectiveness and behavior from an information-theoretic perspective!

1/ Are two #LLMs better than one for equitable cultural alignment? 🌍 We introduce a Multi-Agent Debate framework — where two LLM agents debate the cultural adaptability of a given scenario. #ACL2025 🧵👇

🚨 New Paper 🧵 How effectively do reasoning models reevaluate their thought? We find that: - Models excel at identifying unhelpful thoughts but struggle to recover from them - Smaller models can be more robust - Self-reevaluation ability is far from true meta-cognitive awareness

CFP of the Wordplay 2025 (EMNLP) is live! wordplay-workshop.github.io

RAG and in-context learning are the go-to approaches for integrating new knowledge into LLMs, making inference very inefficient We propose instead 𝗞𝗻𝗼𝘄𝗹𝗲𝗱𝗴𝗲 𝗠𝗼𝗱𝘂𝗹𝗲𝘀 : lightweight LoRA modules trained offline that can match RAG performance without the drawbacks

🎉 Yay, welcome hyunji amy lee -- super excited to have you join us as a postdoc! 🤗 Welcome to our MURGe-Lab + UNC AI + UNC Computer Science family & the beautiful Research Triangle area -- looking forward to the many fun+impactful collaborations together 🔥

PS. FYI, Hyunji (Amy)'s expertise/interests are in retrieval, aligning knowledge modules with LLM's parametric knowledge, and expanding to various modalities, with diverse+extensive work at KAIST, AI2, Adobe, MSR, etc., details at --> amy-hyunji.github.io

🚀 We introduce GrAInS, a gradient-based attribution method for inference-time steering (of both LLMs & VLMs). ✅ Works for both LLMs (+13.2% on TruthfulQA) & VLMs (+8.1% win rate on SPA-VL). ✅ Preserves core abilities (<1% drop on MMLU/MMMU). LLMs & VLMs often fail because

Maybe don't use an LLM for _everything_? Last summer, I got to fiddle again with content diversity Adobe Research Adobe and we showed that agentic pipelines that mix LLM-prompt steps with principled techniques can yield better, more personalized summaries