Zhijun Yin

Two AI systems with nearly identical performance can have drastically different human oversight needs, revealing hidden costs in deployment readiness.

Discrepancies in medication documentation reveal that nearly 90% of clinic notes may still align temporally with structured EHRs, despite low exact agreement.

DRIFT outperforms existing models by effectively integrating treatment interventions into ICU forecasting, achieving lower error rates even when treatment sequences change.

LLM conformity isn't just about sycophancy; it's also significantly driven by a model's own uncertainty, opening new avenues for targeted interventions.