Xinping Lei

Cultural competence in language models is more about pre-training exposure than multilingual fluency, revealing a critical gap in AI's understanding of cultural nuances.

Fine-tuning a model on rigorously synthesized tasks can outperform larger models by leveraging high-fidelity data, achieving a new benchmark in terminal agent performance.