Tag
GiveDirectly's pilot in rural Rwanda paired unconditional cash transfers with a general-purpose AI chatbot, revealing both value as an always-available advisor and critical limitations including language gaps, irrelevant responses, and confidently incorrect answers, raising questions about evaluating models beyond benchmarks.
XLGoBench introduces a synthetic benchmark of algorithmic tasks to detect cross-lingual skill gaps in LLMs, demonstrating persistent gaps across multiple state-of-the-art models.