Hugging Face 推出 Community Evals：別再盲信黑箱排行榜，讓社群來決定模型好壞！

In today's era of rapid AI advancement, major model vendors and research institutions are releasing all manner of "leaderboards" to claim their models surpass the competition. However, the evaluation methodologies, test datasets, and scoring criteria behind these leaderboards are often opaque, and some suffer from serious issues of "data contamination" and metric gaming (Goodhart's Law). To address this pain point, Hugging Face, the leading open-source AI community, officially announced the launch of "Community Evals," declaring: "We will no longer blindly trust black-box leaderboards — instead, we will return control of evaluation to the community."