🚀 Introducing AutoDAN, a method that automatically generates SEMANTICALLY MEANINGFUL #Jailbreak prompts for #redteaming aligned #LLMs . arxiv: arxiv.org/pdf/2310.04451…

I'm grateful for the opportunity to serve as SIGDAT officer. Thank you all for the support and ideas contributing to improving our community. 🥰 sigdat.org/organization

🔥 Unlocking the power of Abstract Meaning Representations, AMRFact generates coherent, factually inconsistent summaries with high error-type coverage to improve the factuality evaluation on abstractive summarization! 📣 Check out our new #NAACL2024🇲🇽work: arxiv.org/abs/2311.09521

Proud of this work where we show event detection can generalize from one epidemic to another by identifying epidemic-related event types (e.g. symptoms) even if the actual mentions (e.g. of symptoms) are distinctive. So training on COVID, we can generalize to Monkeypox! #NAACL24

Excited about training on synthetic data? Different stages of training might need different synthetic data! 🧠💡 Check out our #ICLR2024 paper on Progressive Dataset Distillation (PDD😉) at PS#2 Halle B#9! It tailors synthetic data to each training stage for better performance!

I have to miss #ICLR2024 due to teaching, but would still like to shoutout for our UniversalNER. This is an extremely strong open NER system that provides precise recognition in many domains and for any new entity types (outperforming ChatGPT by 9% on 43 NER datasets across 9

🎳Jailbreak of "Safe" Large Language Models is as simple as a string replacement. 💥Our recent research finds that specifying an output prefix of LLMs will make the jailbreak easy and effective. researchsquare.com/article/rs-438…

Jailbreaking through asking for coherent outputs 🧵📖 Read of the day, day 49: Frustratingly Easy Jailbreak of Large Language Models via Output Prefix Attacks, by Yiwei Wang et al from UCLA Yet another way of breaking through LLM’s defense systems found. This idea is

Professor Nanyun (Violet) Peng Receives NSF CAREER Award Read more: cs.ucla.edu/professor-nany…

🚀Excited to share our latest research on knowledge editing (KE) in large language models! We unveil a novel approach, DeCK, which enhances in-context editing (ICE) by addressing stubborn knowledge that is tough to edit. DeCK boosts ICE performance by up to 219% on MQuAKE! 🚀

Introducing WPO: Enhancing RLHF with Weighted Preference Optimization 🌟 Our new preference optimization method reweights preference data to simulate on-policy preference optimization using off-policy data, combining efficiency with high performance. ✅ up to 5.6% better than

🌟 𝐌𝐮𝐥𝐭𝐢𝐦𝐨𝐝𝐚𝐥 𝐃𝐏𝐎🌟 🔍 DPO over-prioritizes language-only preference 🚀 Introducing mDPO: optimizes image-conditioned preference 🏆 Best 3B MLLM with reduced hallucination, beats LLaVA 7/13B with DPO Collaboration with Microsoft Research huggingface.co/papers/2406.11…

LLM研究者本人による解説記事が出ました。 ai-data-base.com/archives/72359 中国科学院大学（University of Chinese Academy of Sciences）のBaolong Bi氏らによる「頑固な知識」の編集アプローチに関する論文です。

🚨 NEW OPENAI MODEL: o1 “o1 spends more time thinking before it responds. In a qualifying exam for the International Mathematics Olympiad (IMO), GPT-4o correctly solved only 13% of problems, while the reasoning model scored 83%. Their coding abilities were evaluated in

o1-preview is 100x more expensive than GPT-4o mini, costing $15 per million input tokens compared to GPT-4o mini's $0.15.

🚨Vision-Language Models (VLMs) are truly amazing. Ever wonder if their visual and textual "brains" always agree? I am excited to share our latest paper, where we tackle a critical challenge in VLMs, dubbed the 𝐜𝐫𝐨𝐬𝐬-𝐦𝐨𝐝𝐚𝐥𝐢𝐭𝐲 𝐩𝐚𝐫𝐚𝐦𝐞𝐭𝐫𝐢𝐜 𝐤𝐧𝐨𝐰𝐥𝐞𝐝𝐠𝐞

⚙️ Introducing METAL! A multi-agent framework to generate charts that precisely replicate visual details in the reference. 📈 We show that test-time scaling with the multi-agent system can bring 5.2% gain over the current best result on ChatMIMIC! 🌐 metal-chart-generation.github.io

Sharing a new GitHub repository for collecting and sharing papers on the emerging topic of Context Engineering, which has seen broad adoption in industry: github.com/Meirtz/Awesome… A corresponding survey paper is also coming soon. Thanks for reading! ☀️

📄 A Survey of Context Engineering for Large Language Models 🧠 arXiv link: arxiv.org/abs/2507.13334 Context Engineering is the art of generating, acquiring, processing, and managing contextual information for language model agents. 📚 GitHub project: github.com/Meirtz/Awesome…