Brief

A new arXiv paper investigates the efficacy of Large Language Models (LLMs) in annotating data for active learning, specifically for hostility detection in online comments. The study found that LLMs, particularly GPT-5.2 with a two-question interface, can label data at a significantly lower cost than human annotators, achieving comparable or superior performance. However, the research also noted that active learning did not provide a reliable advantage over random sampling when using LLM annotators, and the error structures of different LLMs varied, with some misclassifying economic or border-control discourse. AI

Researchers have developed a new dataset of 1,893 user questions specifically designed for explainable robotics. This dataset, collected from 100 participants, categorizes questions into 12 types, focusing on user expectations for robot capabilities and task execution details. The findings indicate that while users frequently ask about basic task information, they consider questions about hypothetical scenarios and ensuring correct behavior to be the most important. This resource aims to aid in developing better question-answering modules and explanation strategies for human-robot interaction. AI

A new research paper explores the effectiveness of crowdsourcing for detecting audiovisual deepfakes. The study found that while crowd workers are generally good at identifying authentic videos, they frequently miss manipulated content and struggle to accurately pinpoint the type or timestamps of manipulation. Aggregating judgments can improve the detection of authenticity but does not fully address the issue of missed manipulations or the difficulty in identifying specific manipulation types, especially for audio-video combined deepfakes. AI