Cross‑lingual Backdoor Attacks Reveal Vulnerability in Multilingual LLMs

Cross‑lingual backdoor attacks expose a vulnerability in multilingual large language models. Read more: getnews.me/cross-lingual-backdoor-a... #crosslingual #backdoor #multilingualllms

ReCoVeR Reduces Language Confusion in Multilingual LLMs

ReCoVeR, a lightweight vector‑based method, lowers language confusion in multilingual LLMs across 18 languages and three benchmarks while keeping task performance intact. Read more: getnews.me/recover-reduces-language... #recover #multilingualllms

Moderating Quechua Content on Social Media In this project we seek to critically examine how content moderation systems operate in non-English contexts, particularly in indigenous and other languages of the Majority World (i.e., the Global Sou...

I am really excited about @cdt.org's latest research report "Moderating Quechua Content on Social Media."

How does social media manage user generated content in Quechua and how useful are LLMs in this process? cdt.org/insights/mod...

#LLMs #contentmoderation #MultilingualLLMs #IndigenousLanguages