Yingqi Wang

Self-authored verification can lead to catastrophic performance regressions in self-improving agents, but integrating an external audit mechanism can stabilize outcomes.

Current machine translation systems often fail to capture the nuances of culturally-loaded expressions, highlighting a critical gap in their ability to truly understand and translate language.

LLM-based multi-agent systems can neutralize 40% more faults with iterative, closed-loop designs compared to linear workflows, revealing architectural topology as a key factor in robustness.