The Transformer's encoder clearly explained 👇🏻

🤔Modern LLMs are known to support long text, but can they 𝐟𝐮𝐥𝐥𝐲 𝐮𝐭𝐢𝐥𝐢𝐳𝐞 the information available in these texts? 💡Introducing 𝐄𝐓𝐇𝐈𝐂, a new long-context benchmark designed to assess LLMs' ability to leverage the entire given context. arxiv.org/abs/2410.16848

Excited to introduce VideoRAG📽️, which extends the current landscape of RAG systems (that largely focus on text) by holistically leveraging the video corpus, utilizing both textual and visual features. Paper: arxiv.org/abs/2501.05874 Soyeong Jeong Kangsan Kim Sung Ju Hwang

My first first-author paper is accepted to NAACL 2025. Extra special as a huge fan of Breaking Bad (It will be held in Albuquerque NM). Paper: Rationale-Guided Retrieval Augmented Generation (arxiv.org/abs/2411.00300)

🚀 New Announcements 🚀 ⏱︎ how do Large language models (LLMs) handle temporally evolving facts? We uncover Temporal Heads, specialized attention heads within LLMs that are primarily responsible for processing time-sensitive knowledge.

Many of you asked if I could make the 'LLMs as GNNs' lecture more broadly available. 💬🕸️ I'm delighted to share that next Wednesday (26 March) is your chance to tune in, via the GLOW seminar! 🌟 Details in Fabrizio's post:

#NLProc New paper on "evaluation-time scaling", a new dimension to leverage test-time compute! We replicate the test-time scaling behaviors observed in generators (e.g., o1, r1, s1) with evaluators by enforcing to generate additional reasoning tokens. arxiv.org/abs/2503.19877

Forget IDP. Agentic information extraction is here. Stop training models for every doc type. Upstage Information Extract understands and extracts data from any complex document—invoices, resumes, contracts, medical forms—with zero training required. API access is coming

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!

Excited to present 𝐄𝐓𝐇𝐈𝐂 at #NAACL2025 ! Looking forward to open discussions on long-context modeling, evaluation, or anything else :) 🗓️ Friday, May 2, 11:00-12:30 📍 Hall 3, Session K, Poster Session 8

🎉 Our paper has been accepted to ACL 2025 Main Conference! 🎉 We're excited to share our work: "Does Time Have Its Place? Temporal Heads: Where Language Models Recall Time-specific Information" Looking forward to connecting with the NLP community at ACL 2025! #ACL2025

🚀 Small. Powerful. Now with Reasoning. Solar Pro 2 Preview from Up stage hits 70B-class performance with 31B size—especially in Korean. Chat for speed. Reasoning for complexity. One compact model that does it all. 📘 Full details → upstage.ai/blog/en/solar-… 🆓 Free API access

🚀 Introducing Solar Pro2 – A Leap Forward in Reasoning AI 🧠✨ After 4–5 months of intense development, iteration, and testing, I'm proud to share Solar Pro2, our latest and most capable reasoning model to date. #SolarPro2 #Upstage #Reasoning