Diogo Almeida· @CompleteSkeptic · X·· 8 天前AI 评分21
AI 导读
刷榜模型与真正在意可靠性的人打造的模型之间，存在一个关键差异，而从定义上讲，这个差异不会体现在基准测试中 （我们在字符串大语言模型上多少已经学到了这一点……）
正文
there's a critical difference between benchmaxxed models and ones made by people who truly care about reliability that, by definition, doesn't show up on benchmarks
(we've somewhat learned this for string LLMs...)
@julianharris There's something wet-sponge-feeling about the qwen fine tunes that I can't explain compared to Jev在 X 查看被引用的帖子
来源：Diogo Almeida · x.com