Menglin Liu

Chinese model guardrails target coordination rather than ideology—declining even to organize pro-government rallies—yet their sky-high refusal rates collapse under basic adversarial paraphrasing.

Positional bias in LLMs can drastically alter predictions, with accuracy and stability often at odds, challenging our understanding of model reliability in ordinal classification tasks.

JailbreakOPT boosts attack success rates against LLMs while slashing the number of attempts needed to breach safety measures.