Latest work
New Arrival
-
A Survey of Architectural Security in Harnesses for Large Language Model Agents
针对大模型智能体通过工具调用改变真实环境时面临的跨部件安全风险，文章以连接模型与环境的运行支撑层Harness为研究对象，采用定向文献检索与引用追踪，梳理1974---2026年的223项资料。围绕运行职责、威胁描述、保障机制及评价与演化建立分析框架，将七类运行部件与攻击者能力、策略实施位置和证据类
-
ITRAQ and PRM-based quantitative saliva proteomics in gastric cancer: biomarker discovery
This study identified salivary biomarkers S100A8, S100A9, CST4, and CST5 as potential tools for early gastric cancer (GC) detection using iTRAQ and PRM proteomics. These biomarkers showed distinct expression patterns and were linked to disease progression, offering a non-invasive, clinically applicable diagnostic approach.
-
AI Model Evaluation
This study evaluates ChatGPT-4's performance on the Chinese National Civil Service Exam, highlighting its strengths in language comprehension and structured writing while noting limitations in logical reasoning and multi-step problem-solving. Findings underscore its potential in academic assessments and the need for further advancements in AI cognitive capabilities.