Power Consumption Patterns Using Telemetry Data

This collection of recent arXiv abstracts surfaces multiple security-relevant themes. Most critically, an empirical study shows substantial LLM-mediated novice uplift on biosecurity-relevant in-silico biology tasks (novices with LLMs were ~4.16x more accurate and in some cases exceeded expert baselines), indicating meaningful dual-use risk from wider LLM access. A large-scale evaluation of Chinese-language information access (12,161 yes/no queries) reveals substantial misinformation exposure differences across search engines, standalone LLMs, and AI overview modules, implying real-world risk/