ArXiv TLDR

Yangqiu Song

4 papers · Latest:

Computer Vision

Training Long-Context Vision-Language Models Effectively with Generalization Beyond 128K Context

This paper introduces MMProLong, a new recipe for training long-context vision-language models effectively, generalizing beyond 128K context.

2605.13831
Computer Vision

MedHorizon: Towards Long-context Medical Video Understanding in the Wild

MedHorizon introduces a new benchmark for long-context medical video understanding, revealing current MLLMs struggle with sparse evidence retrieval and clinical reasoning.

2605.06537
Computer Vision

Divide-then-Diagnose: Weaving Clinician-Inspired Contexts for Ultra-Long Capsule Endoscopy Videos

This paper introduces a new task, dataset (VideoCAP), and framework (DiCE) for diagnosis-driven summarization of ultra-long capsule endoscopy videos.

2604.21814
Cryptography & Security

Into the Gray Zone: Domain Contexts Can Blur LLM Safety Boundaries

Jargon exploits LLM safety boundaries by leveraging domain contexts, achieving >93% attack success on frontier models, and proposes a policy-guided safeguard.

2604.15717

📬 Weekly AI Paper Digest

Get the top 10 AI/ML arXiv papers from the week — summarized, scored, and delivered to your inbox every Monday.