Privacy dilemmas and opportunities in large language models
Peer-Reviewed Publication
Updates every hour. Last Updated: 24-Dec-2025 14:11 ET (24-Dec-2025 19:11 GMT/UTC)
Survey dissects 5 LLM text-privacy threats in train/inference plus 3 app-centric gaps, urging native safeguards before real-world rollout
Researchers lift FSS from unary to multivariate: two-layer OT-linked binary trees shrink distributed comparison function key size from O(λn²) to O(λn)
MPFToD lifts SOTA 56.3%→61% via three modular tasks and transfers to dialogue acts & more
SSL-powered EFAM: bi-path fusion and channel attention elevate SED accuracy while cutting labeled-data needs
CAS-FEND distills comment knowledge into a content-only student, beating comment-aware rivals with a quarter of comments for timely fake-news catch
A new study published in Microbiome Research Reports reveals that a polyphenol-rich dietary pattern can significantly reduce inflammation and improve gut microbiota composition in adults aged 60 years and older with elevated inflammatory markers. Conducted as part of the MaPLE (Microbiome mAnipulation through Polyphenols for managing Leakiness in the Elderly) trial, this research showed that an eight-week polyphenol-rich diet decreased key inflammatory markers such as IL-6 and CRP, enhanced microbial diversity, and increased beneficial bacteria including Blautia and Dorea. Metabolomic analyses indicated notable shifts in polyphenol-derived metabolites associated with anti-inflammatory effects. These findings highlight the potential of polyphenol-rich diets as personalized nutritional strategies to counteract “inflammaging” and promote healthy aging.
Tsinghua University Press is pleased to announce the official launch of Ocean (www.sciopen.com/journal/3008-1203), an international, peer-reviewed open-access journal dedicated to advancing research in ocean science, technology, and engineering.
Autonomous driving systems increasingly rely on data-driven approaches, yet many still struggle with reasoning, handling rare scenarios, and transparently explaining their actions. A new study introduces DriveMLM, a multi-modal large language model framework that aligns language-based reasoning with structured behavioral planning states, enabling full closed-loop driving in realistic simulators. By integrating multi-view images, LiDAR inputs, traffic rules, and natural-language instructions, DriveMLM generates both driving decisions and human-readable explanations that map directly to vehicle control. The system significantly improves safety, adaptability, and interpretability, demonstrating how large language models (LLMs) can advance the next generation of autonomous driving technology.