Pig butchering and similar interactive online scams lower their victims' defenses by building trust over extended periods of conversation — sometimes spanning weeks or months. These scams have resulted in massive public losses, estimated at over $75 billion in a recent study. In this paper, we describe the motivation, design, implementation, and experience with CHATTERBOX, an LLM-based system that automates long-term engagement with online scammers, making large-scale investigations of their tactics possible. We detail the techniques developed to attract scam attempts, the system architecture and LLM engineering required to convincingly engage with scammers, and the necessary capabilities to satisfy or evade "milestones" in the scammers' workflow.

SceMQA is a novel benchmark for scientific multimodal question answering at the college entrance level, addressing a critical educational phase often overlooked in existing benchmarks. It covers Mathematics, Physics, Chemistry, and Biology with a blend of multiple-choice and free-response formats, providing knowledge points and detailed explanations for every problem. Evaluations of state-of-the-art open- and closed-source MLLMs show only 50–60% accuracy for the strongest models, highlighting the need for further research.

This study integrates large language models with multi-modal analogical reasoning to enhance STEM education: a system that transforms intricate principles in mathematics, physics, and programming into comprehensible visual metaphors, evaluated via a randomized A/B/C test measuring learning gains and motivation.