Jing Shao

Researcher@Shanghai AI Laboratory·中国·OpenReview

研究方向

AI Safety · Multimodal Learning · Representation Learning · Action Recognition and Detection · Scene Understanding · Face Recognition · Face Generation · Crowd Analysis

Jing Shao

ExGRPO: Learning to Reason from Experience

When AI Agents Collude Online: Financial Fraud Risks by Collaborative LLM Agents on Social Platforms

Your Agent May Misevolve: Emergent Risks in Self-evolving LLM Agents

Rethinking Bottlenecks in Safety Fine-Tuning of Vision Language Models

Conditional Advantage Estimation for Reinforcement Learning in Large Reasoning Models

Towards Self-Evolving Agent Benchmarks : Validatable Agent Trajectory via Test-Time Exploration

The Devil behind the mask: An emergent safety vulnerability of Diffusion LLMs

Property-Oriented and Structurally Minimal Feedback for Effective LLM-based Code Refinement

Unintended Harmful Knowledge Elicitation Issue in Large Reasoning Models and a RL Solution

Rethinking Entropy Regularization in Large Reasoning Models

STaR-Attack: A Spatio-Temporal and Narrative Reasoning Attack Framework for Unified Multimodal Understanding and Generation Models

Dive into the Agent Matrix: A Realistic Evaluation of Self‑Replication Risk in LLM Agents

PACEbench: A Framework for Evaluating Practical AI Cyber-Exploitation Capabilities

Taming Masked Diffusion Language Models via Consistency Trajectory Reinforcement Learning with Fewer Decoding Step

SafeAgentBench: A Benchmark for Safe Task Planning of Embodied LLM Agents

REEF: Representation Encoding Fingerprints for Large Language Models

RiOSWorld: Benchmarking the Risk of Multimodal Computer-Use Agents

Systematic Reward Gap Optimization for Mitigating VLM Hallucinations

Demystifying Reasoning Dynamics with Mutual Information: Thinking Tokens are Information Peaks in LLM Reasoning

WorldSimBench: Towards Video Generation Models as World Simulators

WorldSimBench: Towards Video Generation Models as World Simulators

Derail Yourself: Multi-turn LLM Jailbreak Attack through self-discovered clues

SPA-VL: A Comprehensive Safety Preference Alignment Dataset for Vision Language Model

MAS-GPT: Training LLMs to Build LLM-based Multi-Agent Systems

Benchmarking Ethics in Text-to-Image Models: A Holistic Dataset and Evaluator for Fairness, Toxicity, and Privacy

OASIS: Open Agents Social Interaction Simulations on a Large Scale

DEAN: Deactivating the Coupled Neurons to Mitigate Fairness-Privacy Conflicts in Large Language Models

AD-H: Autonomous Driving with Hierarchical Agents