不便外发的段落 | Midjourney

3.2 Midjourney 中的内容审核和审查 | Content moderation and censorship in Midjourney

文中可能包含无法从中国内地访问的链接。

2023年3月，大卫·霍尔兹表示，为防止中国政府可能因为习近平的讽刺图像而封锁Midjourney，Midjourney已经阻止了习近平图像的生成。霍尔兹说：“中国人使用这项技术的能力比你生成讽刺作品的能力更重要。”[18]

Prior to May 2023, Midjourney implemented a moderation mechanism predicated on a banned word system. This method prohibited the use of language associated with explicit content, such as sexual or pornographic themes, as well as extreme violence. Moreover, the system also banned certain individual words, including those of religious and political figures, such as Allah or Xi Jinping. This practice occasionally stirred controversy due to perceived instances of censorship within the Midjourney platform.[47][48]
【参考译文】2023 年 5 月之前，Midjourney 实施了基于禁词系统的审核机制。这种方法禁止使用与露骨内容相关的语言，例如性或色情主题，以及极端暴力。此外，该系统还禁止使用某些个别词语，包括宗教和政治人物的词语，例如真主或习近平。由于 Midjourney 平台内存在审查制度，这种做法偶尔会引发争议。[47][48]

Commencing in May 2023, with subsequent updates post version 5, Midjourney transitioned to an AI-powered content moderation system. This advanced mechanism allowed for a more nuanced interpretation of user prompts by analyzing them in their entirety. It consequently facilitated the context-dependent use of words that had previously been prohibited. For instance, users can now prompt the AI to generate a portrait of Xi Jinping. At the same time, the system will prevent the generation of contentious images, such as depictions of global leaders, including Xi Jinping, in situations of arrest.[49]
【参考译文】从 2023 年 5 月开始，随着版本 5 的后续更新，Midjourney 过渡到由人工智能驱动的内容审核系统。这种先进的机制允许通过对用户提示进行整体分析来更细致地解释用户提示。因此，它促进了以前被禁止的词语的上下文相关使用。例如，用户现在可以提示人工智能生成习近平的肖像。同时，该系统将防止生成有争议的图像，例如在被捕情况下描绘包括习近平在内的全球领导人。[49]