flâneur

Vedaangh Rungta

vedaangh.com · 1,410 words · saved by 1 readers

Personal Website

← Back Introduction Chinese open-weight and open-source–style LLMs are now easy to download, fine-tune, and self-host, and they dominate the open-source LLM landscape, where Western alternatives are few and far between. There has been little public work investigating the political alignment and institutional allegiances of these models, and the risks that may arise from deploying them in Western contexts. To make those risks more legible, we evaluated six Chinese frontier models across six behavioral evaluations. Models were evaluated on: (1) answering questions on matters sensitive to the CCP

related reading