Leyang Cui

@yafuly congratulation!

Excited to have 6 papers accepted to #ICLR2026, all around reasoning, RL, and multimodal understanding: 📌ExGRPO: Learning to Reason from Prior Successes 📌Diversity-Incentivized Exploration for Versatile Reasoning 📌Conditional Advantage Estimation for Reinforcement Learning in Large Reasoning Models 📌Spotlight on Token Perception for Multimodal RL 📌Revisual-R1: Advancing Multimodal Reasoning from Optimized Cold Start to Staged RL 📌FrameThinker: Learning to Think with Long Videos via Multi-Turn Frame Spotlighting 💻All works are open-sourced — welcome discussions, feedback, and collaborations! Huge thanks to all collaborators. Looking forward to great discussions at ICLR! @iclr_conf #iclr

@yafuly Congratulation!!!

🎉 Excited to share that our paper “Siren’s Song in the AI Ocean: A Survey on Hallucination in Large Language Models” — the first comprehensive survey on hallucinations in LLMs — has been accepted to Computational Linguistics (CL) and will be presented at EMNLP 2025! 🌊🤖@emnlpmeeting Since the preprint release, it has drawn remarkable attention — now cited over 1,600 times, reflecting the community’s deep interest in understanding and mitigating hallucination in LLMs. Looking forward to meeting everyone at #EMNLP2025 (Hall C, Session 15) to discuss interpretability, transparency, and trustworthy reasoning in large language models. #LLM #AIResearch #Hallucination #ComputationalLinguistics #EMNLP

Echoing the great work from @dongkeun_yoon, also share our updated preprint! 🧐 Do reasoning models verbalize their confidence better than instruct models? 🧐 🧐 Does RL provide additional benefits? 🧐 We explore this using a series of instruct and reasoning models... [1/n]

🧐Do LLMs solve math problems by truly understanding and applying math knowledge🔦📝, or do they rely on 𝐬𝐮𝐩𝐞𝐫𝐟𝐢𝐜𝐢𝐚𝐥 𝐩𝐚𝐭𝐭𝐞𝐫𝐧𝐬 or potential 𝐝𝐚𝐭𝐚 𝐥𝐞𝐚𝐤𝐚𝐠𝐞? 📣 Excited to announce🌟𝐆𝐒𝐌-𝐏𝐥𝐮𝐬 🌟, an adversarial dataset for the math word problems.

📢 New paper! 📄 A Comprehensive Study of GPT-4V's Multimodal Capabilities in Medical Imaging. arxiv.org/abs/2310.20381 🌟First quantitative evaluation on publicly available benchmarks. 🌟Covers Radiology Report Generation, Medical Visual Question Answering and Visual Grounding. 🌟Expert evaluations with actual physicians for in-depth insights! Stay tuned for groundbreaking findings! 🔍 #MedicalImaging #GPT4V #LLM

Introducing 🦜Macaw-LLM🦜: A pioneering multi-modal language model integrating images🖼️, videos🎥, audios🔊 & text📝! Macaw-LLM boasts simple & fast alignment to LLM embeddings , and one-stage instruction fine-tuning🛠️. Built on the foundations of CLIP, Whisper, and LLaMA, our project is in its early stages, but we're excited to share our work with the community. Check it out on GitHub: github.com/lyuchenyang/Ma…. 🤗Your feedback and contributions are welcome as we continue to explore and innovate in this exciting area of study! 🌏