📖 Excited to present our novel long-story generation framework, Collective Critics for Creative Story Generation (CritiCS)! Previous long-story generation research mainly focused on coherence, overlooking creative and captivating storytelling. However, what we truly want is to generate stories that captivate readers. Therefore, we developed CritiCS—a framework designed to foster creativity and produce captivating long-form stories through a collective critique mechanism. Thanks to @khsquared and the amazing Language & Intelligence Lab family for their invaluable insights and continuous support!

🎉Excited to introduce our new paper: Unbounded: A Generative Game of Character Life Simulation! We build a game of character life simulation that is fully encapsulated in generative models. 🌟We achieve this with: ▶️ A specialized, distilled LLM that dynamically generates game mechanics, narratives, and character interactions in real-time. ▶️ A dynamic regional IP-Adapter for vision models that ensures consistent yet flexible visual generation of a character across multiple environments. 🧵

💭🔎 Introducing EvalPlanner – a method to train a Thinking-LLM-as-a-Judge that learns to generate planning & reasoning CoTs for evaluation. Strong performance on RewardBench, RM-Bench, JudgeBench & FollowBenchEval. Paper 📄: arxiv.org/abs/2501.18099

🎉Congratulations to Prof. @mohitban47 on being named a 2025 @RealAAAI Fellow for "significant contributions to multimodal AI foundations & faithful language generation and summarization."👏 16 fellows chosen worldwide by cmte of 9 past fellows & ex-pres: aaai.org/about-aaai/aaai-awa…

🎉 Congratulations to Prof. @mohitban47 for receiving the Presidential #PECASE Award by @WhiteHouse, which is the highest honor bestowed by US govt. on outstanding scientists/engineers who show exceptional potential for leadership early in their careers! whitehouse.gov/ostp/news-upd…

Join the Dept of #ComputerScience on December 6th at 2:30 PM to welcome Mohit Bansal, UNC Chapel Hill, who will lecture on "Planning Agents for Collaborative Reasoning and Multimodal Generation." Learn more here: tinyurl.com/jkmex3e9 @AI_SBU @CEASSBU @stonybrooku

🚨 Excited to announce “Teaching Models to Balance Resisting and Accepting Persuasion” LLMs need to be able to reject negative persuasion (e.g. misinfo, jailbreaking, etc)... BUT we argue that they also need to accept positive persuasion (e.g. when they are wrong/unsafe) to be helpful! We take a 1st step in balancing: 1⃣ defending against negative persuasion (makes answers worse) with 2⃣ accepting positive persuasion (makes answers better), training LLMs to accept persuasion when appropriate. Our multi-agent method, Persuasion-Balanced Training (PBT) recursively creates RLHF training data from generated dialogue trees and then trains models on positive/negative persuasion. Across three models of varying sizes, PBT -- improves resistance to misinformation -- reduces flipflopping -- obtains best performance on balanced data -- makes models better teammates/robust to ordering in multi-agent discussion When pairing 2 LLMs in a multi-agent debate, we find that pairs generally have a lot of order-dependence. Depending on whether the stronger or weaker model goes first, the team lands on the stronger or weaker answer. PBT reduces this variability and improves team performance. 🧵👇

🚨 We’re thrilled to announce our exciting keynotes for #EMNLP2024 2024.emnlp.org/program/keyno… 1⃣  11/12: Percy Liang: open-source science 2⃣  11/13: Anca Dragan: safety alignment 3⃣  11/14: Tom Griffiths: Bayes AI See you in Miami! 🌴 @percyliang @ancadianadragan  @cocosci_lab

Excited to share that "Mixed-Session Conversation with Egocentric Memory" has been accepted to #EMNLP2024 Findings🎉🎉🎉 Special thanks to @kty4119 and @khsquared for their collaboration! arxiv.org/abs/2410.02503

🚨 Introducing Adapt-♾: Scalable Lifelong Multimodal Instruction Tuning via Dynamic Data Selection! ▶️ New multimodal instruction tuning datasets are continuously released, often containing redundant/similar content or targeting highly varied skills (i.e., tasks). ▶️ How can we enable scalable, lifelong instruction tuning for MLLMs, where a temporal stream of multi-task, multimodal instruction-tuning datasets are continually added to the existing training pool? ▶️ We present Adapt-♾, a scalable and adaptive data selection strategy that facilitates the effective learning of MLLM on new skills while reinforcing previously learned skills over time. 📖: arxiv.org/abs/2410.10636 Thread 🧵👇

Congratulations @sumukx @AbhinavChinta10 @VaibhavS2099 @dilekhakkanitur looking forward to your presentation in Miami 🎉