看完了这个推文，后背真的有点发凉。 我们经常浸润在AI赛道以为自己有一双能一眼看穿“AI味”的火眼金睛： 但现实直接给了我们一记响亮的耳光——我们自以为是的“鉴赏力”，不过是在经验主义的盲区里打转。 更讽刺的是，当两幅作品放在一起让你二选一时，你满脑子都是“找出那个真的、揪出那个假的”，甚至自信满满地分析笔刷与肌理，却根本没想过：如果两幅全都是AI呢？ 这才是最让我感到无力和警惕的地方。 在多模态混杂的对比下，AI的拟真与反向伪装（甚至故意模仿人类的笔触与破绽）早已超越了肉眼的极限。 我们不是在鉴别真伪，而是在把自己的思维偏见投射在画面上。 我们不能再盲目的自行了。面对现在的AI，放下那种“我一眼就能看出来”的傲慢吧。 我们赖以判断真实的标准正在被成体系地瓦解，而自我认知的更新速度，可能远远跟不上它进化的脚步。 The results are in. The Ariel pencil test was made with #AI. The color version is the original movie footage. I left both versions unlabeled because I wanted to see how people’s assumptions about AI would affect their judgment. While some of you spotted it right away (I even left a few clues, like the incorrect peg holes and random annotations), the majority of you criticized the color version thinking it was AI and praised the pencil test believing it was #GlenKeane’s original work. That’s the bias against AI I wanted to highlight. People were pointing out supposed AI flaws in the original Disney animation, while praising the AI-generated pencil test for its human touch. If you preferred the pencil version, you can still prefer it. The video hasn’t changed. Only what you know about it has.