📄 中文摘要
AI 基准测试报告了大型语言模型（LLMs）在特定任务上的表现，但对其驱动性能的基本能力提供的洞见有限。这些基准测试无法解释失败或可靠预测新任务的结果。为了解决这一问题，微软研究人员与普林斯顿大学和瓦伦西亚理工大学合作，推出了 ADeLe。该系统旨在预测和解释 AI 在不同任务中的表现，通过分析模型的能力和局限性，为研究人员和开发者提供更深入的理解，从而提升 AI 系统的可靠性和适应性。
ADeLe：预测和解释 AI 在各任务中的表现
出处: ADeLe: Predicting and explaining AI performance across tasks
发布: 2026年4月1日
📄 English Summary
ADeLe: Predicting and explaining AI performance across tasks
AI benchmarks typically report the performance of large language models (LLMs) on specific tasks but offer limited insights into the underlying capabilities that drive their performance. They fail to explain failures or reliably predict outcomes on new tasks. To address this gap, researchers from Microsoft, in collaboration with Princeton University and Universitat Politècnica de València, have introduced ADeLe. This system aims to predict and explain AI performance across various tasks by analyzing the capabilities and limitations of models. It provides researchers and developers with a deeper understanding, enhancing the reliability and adaptability of AI systems.