🔎 AIGC Detection

🤖 AAAI2026 · 3 paper notes

ActiShade: Activating Overshadowed Knowledge to Guide Multi-Hop Reasoning in Large Language Models

This paper proposes ActiShade, a framework that detects "overshadowed" key phrases in LLM multi-hop reasoning via Gaussian noise perturbation, retrieves supplementary documents using a customized contrastive learning retriever, and iteratively reformulates queries to mitigate error accumulation caused by knowledge overshadowing. ActiShade significantly outperforms DRAGIN and other state-of-the-art methods on HotpotQA, 2WikiMQA, and MuSiQue.

BAID: A Benchmark for Bias Assessment of AI Detectors

This paper introduces the BAID benchmark (208K sample pairs covering 7 bias dimensions and 41 subgroups) to systematically evaluate the fairness of 4 open-source AI text detectors across demographic and linguistic subgroups, revealing significant recall disparities for dialect, informal English, and minority group texts.

Optimized Algorithms for Text Clustering with LLM-Generated Constraints

This paper proposes the LSCK-HC framework, which leverages LLMs to generate set-form must-link/cannot-link constraints (as opposed to traditional pairwise constraints), coupled with a penalty-based local search clustering algorithm. The approach achieves clustering accuracy comparable to SOTA on five short-text datasets while reducing the number of LLM queries by more than 20×.