I ran the same AI task on a weaker model and a stronger one. The weaker one won.
中文摘要
由于优化变量错误，较弱的AI模型在任务中击败了更强的模型，揭示了基准测试的局限性。
English Summary
A weaker AI model outperformed a stronger one because current benchmarks and optimization targets are misleading.
原文节选
Everyone is optimizing the wrong variable, and the benchmarks are never going to tell you that. Continue reading on Medium »