Quang Minh Nguyen

The phrasing of belief expressions can swing LLM accuracy from a +50% boost to a -14% drop, revealing a critical vulnerability in how models navigate user beliefs and facts.

LLMs struggle to self-report adversarial influences, with a mere 27.3% accuracy in recognizing compromised outputs, raising concerns about their reliability in safety-critical applications.

Uncertainty estimates from LLMs often fail when they're needed most – in low-information scenarios – but a simple post-hoc calibration can fix them.