Because with, without that, I mean, percent evaluation accuracy is fine, but that means that three in 10 interactions are failing and you can't tell necessarily.
来源:Explaining Eval Engineering | Galileo's Vikram Chatterji · Chain of Thought — host and credited guests章 t-21段 sentence:60ce5faea27b3e544b52f7adChain of Thought — chainofthought.show; https://chainofthought.show/podcast/46-explaining-eval-engineering-galileos-vikram-chatterji/; CC BY 4.0 https://creativecommons.org/licenses/by/4.0/ ; selection/segmentation by SharedEnglishCorpus. Audio not licensed.