Anthropic 的 Claude 模型在细节把控上被认为领先
精选理由
一位博主观察到 Claude 对细节特别较真，猜测这是对齐训练的结果，观点挺有意思。
一篇 X 帖子比较了各家模型在细致度上的表现。作者最初以为差距来自专有数据而非注意力或能力，后来认为根源在对齐训练。帖子提到 Claude 模型版本为 V4.1F（而非 4.1F-1017），并称 Claude 非常害怕遗漏任何细节。
原文 · Teortaxes
There's something else that Anthropic's models lead on here At first I thought to say that this is about proprietary data, not attention/capability. But now, I think it's about alignment. Claudes are *really* afraid to miss any detail. Btw this is V4.1F (*not* 4.1F-1017). https://t.co/8MlznT6t1U
