作者位次 #6/6NeurIPS 2026Poster
MLLMs Fail to Refuse when Using Tools Agentically
Rikiya Takehi · Ryo Hachiuma · Shaona Ghosh · Dan Zhao · Frank Wang · Yusuke Hirota
智能体与工具OpenReview
首次获取该候选人的完整档案及 StartLux 评估共 20 积分；每位用户单独解锁，之后查看免费。主动刷新 20 积分。
实时查找电子邮箱，费用不超过20积分；未找到邮箱不扣费。
Rikiya Takehi · Ryo Hachiuma · Shaona Ghosh · Dan Zhao · Frank Wang · Yusuke Hirota
Ryo Hachiuma · Arun G Zachariah · Barnaby Simkin · Jibin Varghese · Fu-En Yang · Chi-Pin Huang · Frank Wang · Yusuke Hirota
Na Min An · Yoonna Jang · Yusuke Hirota · Ryo Hachiuma · Isabelle Augenstein · Hyunjung Shim
Zhen Wan · Chao-Han Huck Yang · Jinchuan Tian · Hanrong Ye · Ankita Pasad · Szu-Wei Fu · Arushi Goel · Ryo Hachiuma · Shizhe Diao · Kunal Dhawan · Sreyan Ghosh · Yusuke Hirota · Zhehuai Chen · Rafael Valle · Chenhui Chu · Shinji Watanabe · Boris Ginsburg · Yu-Chiang Frank Wang
We introduce a voice-agentic framework that learns one critical omni-understanding skill: knowing when to trust itself versus when to consult external audio perception. Our work is motivated by a crucial yet counterintuitive finding: naively fine-tuning an omni-model on both speech recognition an...