Summary in Plain Language
This article starts with a rather dramatic incident: a well-known American investor, Druckenmiller, published an op-ed in The Wall Street Journal that was immediately recognized by readers as having obvious signs of being written by AI. He was eventually forced to admit that the article was actually AI-generated. The key clue that exposed this was the frequent use of the phrase “not…but…” throughout the article. By examining this detail, the article reveals a phenomenon that many people are unaware of: in the past two years, the use of this type of profound sentence structure (“not…but…”) in official corporate documents in the United States has increased by four times, becoming a hallmark of AI-generated content. Essentially, AI has quietly infiltrated all types of text that require serious reasoning and credibility, acting like a Trojan horse that erodes our trust in the idea that “the content under a signature represents the true thoughts of the author.” Nowadays, content platforms and regulatory authorities around the world are taking steps to address this issue of trust erosion.
---
Detailed Analysis
1. How Did “Not…But…” Become a Signature of AI-Generated Content?
Three years ago, the phrase “not…but…” was a powerful tool for writers to overturn common misconceptions and present their core arguments in a clear and persuasive way. For example, Kennedy’s famous speech about landing on the moon used this phrase to convey the significance of the mission, making it a timeless classic. However, its meaning has now changed. According to Barron’s Weekly, the frequency of this phrase in official corporate documents in the U.S. has doubled in the past two years, exceeding the average of previous years by more than four times. Many writers are now hesitant to use it, fearing that readers will think, “This wasn’t written by a human; it must be AI.” It’s like the Trojans, who, after being infected by the Trojan horse, were wary of anything coming from the sea, fearing it contained enemies. Similarly, readers now see “not…but…” phrases with suspicion, wondering if the content truly reflects genuine human thought.
2. Why Does Human-Generated Content and AI-Generated Content Sound So Different When Using the Same Phrase?
When humans use “not…but…”, it’s usually done with a specific purpose, such as correcting a widespread misunderstanding. For instance, if everyone around you thinks that “slacking off at work is beneficial,” you might use this phrase to point out that it’s not the company that benefits but you losing time that could have been used for learning. However, when AI uses this phrase, it often lacks context and purpose. It may create false statements without any basis in reality. For example, if you say “I’m having beef noodles for dinner,” AI might respond with, “You’re not just having beef noodles; you’re rewarding yourself after a busy day.” Such responses sound insincere and lack any real depth.
3. AI’s Use of “Not…But…” Is a Result of the Scoring System
AI’s preference for this phrase is not accidental but a product of its training process. AI learns from high-quality examples, such as famous speeches and authoritative reports, which often contain this phrase. When humans evaluate AI-generated content, they tend to give higher scores to those with clear logic and solid arguments. To achieve higher scores, AI automatically includes this phrase, essentially “cheating” the scoring system. Moreover, since AI writes without a real understanding of the context, it often uses this phrase indiscriminately, leading to an overuse of this phrase that sounds fake.
4. This Is More Than a Simple Overuse of Phrases; It’s a Trojan Horse in the Content Industry
This issue is far more serious than just a linguistic flaw. It undermines the foundation of trust in the content industry. In the past, articles signed by financial experts were seen as the result of years of experience and thoughtful judgment. Now, with AI capable of generating such articles in seconds, it’s difficult to distinguish between genuine human thought and AI-generated nonsense. This challenges our belief that signed content reflects the author’s real opinions, potentially leading us to question the credibility of all authoritative content.
5. Global Efforts to Restore Trust
Industry and regulatory bodies are taking action to address this issue. Platforms like Substack and Quora have implemented AI content detection tools, and LinkedIn has added a button for users to report suspected AI-generated content. The EU’s Artificial Intelligence Act requires that all AI-generated content clearly indicate its origin. The goal is not to ban AI from writing but to ensure that its use is transparent and that the responsibility for the content’s credibility is clearly defined. After all, AI cannot be held accountable for the reliability of the content, and the real author must bear the consequences.
In summary, the widespread use of “not…but…” by AI is not just a matter of poor writing but a serious threat to the trust in the content industry. It’s a Trojan horse that erodes our trust in the authenticity of information, and global efforts are underway to repair this gap in trust.