1
These LLMs are the best at resisting Russian propaganda
最新基准测试显示，Anthropic的Claude模型在抵制俄罗斯宣传方面表现最优，Opus 4.7获得“杰出”评级。
Estonian government benchmark shows how dozens of models combat Russia's "strategic narratives."
2
Long Live Fine-Tuning: Task-Specific Transformers Outperform Zero-Shot LLMs for Misinformation Response Classification on Reddit
最新研究证实：针对特定任务微调的Transformer在虚假信息分类上全面碾压零样本LLM，颠覆“大模型万能”认知。
arXiv:2606.04274v1 Announce Type: new Abstract: As large language models (LLMs) become default tools for online information verification, an implicit …
3
ExTax: Explainable Disinformation Detection via Persuasion, Emotion, and Narrative Role Taxonomies
从说服、情感与叙事角色三大维度，构建可解释的虚假信息检测框架，直击LLM生成的高流畅性谎言。
arXiv:2605.27045v1 Announce Type: new Abstract: The democratization of LLMs has accelerated the generation and circulation of highly fluent disinforma…
4
Are Rationales Necessary and Sufficient? Tuning LLMs for Explainable Misinformation Detection
KDD 2026研究：LLM调优用于可解释虚假信息检测，探讨解释（rationale）的必要性与充分性
arXiv:2605.19285v1 Announce Type: new Abstract: The rapid spread of misinformation on social media platforms has become a formidable challenge. To mit…