Haobo Wang

LabRobFail reveals that a specialized vision-language model can boost robotic task success rates by up to 20% through enhanced failure understanding.

Dense retrievers miss the mark on answerability, dropping QA performance drastically despite high semantic relevance, revealing a critical oversight in RAG systems.

Over 57% of training samples suffer from suppressed criteria, but a new self-distillation method can reverse this trend and enhance performance in RL tasks.

OPSD can backfire, leading to rote memorization instead of enhancing reasoning, but a novel decomposition approach reveals a path to meaningful improvements.

Adversaries can exploit structural vulnerabilities in function-calling LLMs to bypass safety measures, achieving high success rates with minimal effort.

ISPO reduces critical reasoning failures in RLVR by transforming reward structures, leading to superior performance on complex reasoning tasks.

SkillComposer enables language models to self-evolve skills in real-time, achieving up to +4.5 improvements on agent tasks compared to larger models.

OPRD closes the performance gap between student and teacher models while training 1.44x faster and using 54% less memory than traditional methods.

Automating scientific instruments is now as easy as mimicking a human using the GUI, unlocking faster and more reproducible research.

Smaller reasoning models can achieve both higher accuracy and shorter reasoning chains by adaptively penalizing unnecessary reflections and coordinating length penalties with problem complexity.