最新数据表明，文本、音频和视频的生产成本已大幅坍塌，导致合成信息激增；据编码审查网站 Graphite 估算，去年在互联网上发布的所有文章中，大约有一半是由人工智能生成的。这样的泛滥使“内容农场”能够以接近零的边际成本，大规模生产模仿正规媒体的欺骗性愤怒诱饵。

那些曾经用来区分可信与不可信信息的传统高成本信号——例如声誉良好的报纸品牌、高质量写作或逼真图像——正在被侵蚀，因为人工智能可以低成本地伪造它们。随着这些信号失效，信息市场正从“分离均衡”滑向“混同均衡”，在这里诚实与不诚实的生产者在统计上变得不可区分，信任、注意力和付费意愿都在下滑。

作为回应，人们提出了新的高成本验证体系，包括为媒体文件建立类似密码学的托管链、依赖高额资本投入的纸质发行，以及在教育中增加手写“工作量证明”式考试。通过重新引入昂贵且难以伪造的信号——类似孔雀的尾巴或侧斑蜥平衡的求偶策略——市场也许能够恢复对准确信息的激励，并把不诚实行为重新压回较小且可持续的少数。

I'm thinking about the appropriate word count for my response. I want to keep it safe, aiming for around 180-220 words, but I can limit myself to a max of 2 sentences per paragraph. My plan is to create three paragraphs with roughly 50-60 words each, focusing on essential numerical data and trends like the fact that about half of internet articles last year were AI-generated. I also want to analyze the context while keeping it concise.

I want to emphasize the share of AI-generated content and the low production costs. Recent data suggests that the cost of producing text, audio, and video has dramatically dropped, leading to a surge in synthetic information. According to Graphite, roughly half of all internet articles published last year were AI-generated. This situation allows "content farms" to mass-produce deceptive rage-bait that closely resembles legitimate outlets at near-zero costs. Making sure to count words quickly is essential here too!

I’m focusing on how traditional costly signals that once distinguished trustworthy information are now eroding due to AI's ability to imitate them cheaply. These signals include reputable newspaper brands, high-quality writing, and realistic imagery. As these signals fail, the market shifts from a separating equilibrium to a pooling equilibrium, making honest and dishonest producers indistinguishable. This decline affects trust, attention, and willingness to pay, resulting in a less reliable information landscape overall.

Source: AI misinformation may have paradoxical consequences

Subtitle: To understand why, consider the side-blotched lizard

Dateline: 12月 04, 2025 05:40 上午