🤔Can you really trust what your AI creates? Meet Silent Branding Attack—a sneaky, no-trigger-needed method to secretly poison text-to-image models! 👀 Test yourself: Can you spot which images below have been silently branded?

🚨 Our new CVPR paper on AI Safety is out: Subtle poisoning can quietly make AI models embed hidden logos—without you knowing or prompting!

SWE tip: The importance of software design is higher than ever, given how well AI can code. Highly recommend checking out the following guide:

Paper2Code Automating Code Generation from Scientific Papers in Machine Learning

So excited to drop PaperCoder, a multi-agent LLM system that turns ML papers into full codebases. It looks like this:📄 (papers) → 🧠 (planning) → 🛠️ (full repos), all powered by 🤖. Big thanks to AK for the shoutout! Paper: arxiv.org/abs/2504.17192

🟢 Announcing the #NAACL2025 Award Winners! The Best Paper and Best Theme Paper winners will present at our closing session 2025.naacl.org/blog/best-pape…

Paper2Code PaperCoder is a multi-agent LLM system that transforms a paper into code repository. It follows a three-stage pipeline: planning, analysis, and code generation, each handled by specialized agents. 100% open-source.

Thrilled to be part of this BiGGen Bench project, which received the Best Paper Award at NAACL 2025 (NAACL HLT 2027). Huge thanks and congratulations to Seungone (Seungone Kim) and all the co-authors! Paper: arxiv.org/abs/2406.05761

One year ago, we introduced ResearchAgent, a system for automatic research idea generation. Since then, the field (e.g., AI Scientist) has rapidly evolved, building on the foundation we laid. Excited to finally present ResearchAgent as an NAACL oral this Wednesday at 11:30 AM!

UniversalRAG: Retrieval-Augmented Generation over Multiple Corpora with Diverse Modalities and Granularities Woongyeong Yeo et al. introduce a RAG system that retrieves from multiple corpora with varying modalities and granularities. 📝arxiv.org/abs/2504.20734 👨🏽‍💻universalrag.github.io

Thanks for highlighting UniversalRAG — a novel RAG framework that adaptively selects text📚, image📸, or video📹 based on each query, making RAG more flexible and effective. Paper: arxiv.org/abs/2504.20734 Woongyeong Yeo Kangsan Kim Jinheon Baek Sung Ju Hwang

Universal RAG RAG is dead, they said. Then you see papers like this and it gives you a better understanding of the opportunities and challenges ahead. Lots of great ideas in this paper. I've summarized a few below:

Introducing 🌌UniversalRAG — a novel RAG framework that adaptively retrieves from multimodal corpora (📕text, 🖼image, 🎥video) at the right level of granularity, from short snippets to long content. Huge thanks to elvis for sharing our work! Paper: arxiv.org/abs/2504.20734

So excited to share that five papers have been accepted to #ACL2025 🎉 Huge thanks to all my amazing collaborators. I am especially grateful that all the internship projects (which I have worked on during my PhD journey) have all found their way to publication. Its rewarding. 😊

Drag-and-Drop LLMs Zero-Shot Prompt-to-Weights

+1 for "context engineering" over "prompt engineering". People associate prompts with short task descriptions you'd give an LLM in your day-to-day use. When in every industrial-strength LLM app, context engineering is the delicate art and science of filling the context window

Are AI scientists already better than human researchers? We recruited 43 PhD students to spend 3 months executing research ideas proposed by an LLM agent vs human experts. Main finding: LLM ideas result in worse projects than human ideas.

Google presents Performance Prediction for Large Systems via Text-to-Text Regression

In the 95th session of #MultimodalWeekly, we have an exciting research presentation on VideoRAG and two hackathon projects built with the TwelveLabs (twelvelabs.io) API.

✅ Soyeong Jeong, Kangsan Kim, Jinheon Baek, and Sung Ju Hwang from KAIST AI will present VideoRAG - a framework that not only dynamically retrieves videos based on their relevance with queries but also utilizes visual & textual information. x.com/jinheonbaek/st…