8 papers
REA: From Correct LLM Judgments to Safe Decision Delivery
Jingxiang Meng, Aofan Liu, Yaoshuai Ma +2
Generative safety guards must classify requests correctly and return outputs that applications can parse. Reasoning guards can improve classification accuracy while producing outpu…
CheatBench: Measuring Reward Gaming in AI Agents
Long Phan, Stephen K. Yang, Jason J. Lim +10
Reinforcement learning has helped AI agents solve increasingly difficult tasks, but high rewards do not always reflect the work users intended. In recent incidents and controlled e…
Route-Align-Verify for Functional Correctness in Code Generation
Erxue Zhou, Jingxiang Meng, Aofan Liu
Large language models (LLMs) have substantially improved code generation, yet achieving strong functional correctness remains difficult, especially for heterogeneous programming ta…
CURE: Local Uncertainty Repair for Block-Parallel Speculative Decoding
Aofan Liu, Jingxiang Meng, Fangxin Liu +1
Speculative decoding mitigates the latency of sequential generation in autoregressive Large Language Models (LLMs) by interleaving draft generation with target verification. Howeve…
TaPR: Test-Aware Policy Refinement for Feedback-Conditioned Code Generation
Aofan Liu, Jingxiang Meng, Fangxin Liu +1
Multi-turn code agents rely on execution feedback to repair incorrect programs, yet standard reinforcement learning paradigms optimize and evaluate policy performance primarily usi…
Paraphrase-Induced Output-Mode Collapse: When LLMs Break Character Under Semantically Equivalent Inputs
Aofan Liu, Jingxiang Meng
When the substantive content of a request is rewritten, do large language models still answer in the format the original task asked for? We find that they often do not, even at tem…
Self-Correction as Feedback Control: Error Dynamics, Stability Thresholds, and Prompt Interventions in LLMs
Aofan Liu, Jingxiang Meng
Iterative self-correction is increasingly deployed in agentic LLM systems, yet whether repeated refinement improves or degrades performance remains inconsistent across models. We r…
Financial black swans driven by ultrafast machine ecology
Neil Johnson, Guannan Zhao, Eric Hunsader +4
Society's drive toward ever faster socio-technical systems, means that there is an urgent need to understand the threat from 'black swan' extreme events that might emerge. On 6 May…