The consumption of AI-generated content at scale

Summary

Introduction A few days ago, I read this tweet: “I might be going insane because half of what I read now sounds like ChatGPT.” In a follow-up, he screenshotted a blog post and said “Like come on.” He was pattern-matching on AI and couldn’t turn it off. I’ve felt this too. There’s a frustration I can’t quite shake when consuming content now—a feeling of being stuck in a loop where everything sounds the same. I’ll read a blog post, skim some code, review a document, and something feels off. Not necessarily wrong, but homogeneous. Like I’ve seen this exact structure, these exact words, a thousand times before. And maybe I have. In this post, I want to explore what I’ve been feeling as a consumer of information in the AI era, and then reflect on it as a researcher. I’ll start by describing two distinct ways my ability to process information has eroded: first, a signal degradation problem where AI has “cried wolf” so often that I’ve stopped noticing the devices meant to help me understand; and second, a verification problem where the ease of generating plausible content has outpaced—and eroded—my ability to verify it. Then, I’ll discuss why this matters—what’s at stake when consumers can’t notice errors or distinguish quality. Finally, I’ll share what I’m currently thinking about how to deal with it. I’ll sketch two threads of thought: one on building systems that capture the why behind the techniques they use, and one on grounding AI confidence in verified human experience rather than simply returning confident-sounding text. Signal Degradation: The AI That Cried Wolf AI has overused the tools designed to aid human comprehension to the point where I’ve stopped noticing them. In complex domains, we’ve developed tools to help humans process information. For example, in writing, a metaphor compresses a complex idea into something the reader already understands. The metaphor is useful precisely because the underlying idea is hard to communicate directly. For example, descri...