🔍 The best guide for detecting AI-generated text

Endless posts on "How to tell if a text was generated by AI" mostly repeat the same favorite AI structures (and, in the worst cases, are written by chatbots themselves).

But unexpectedly, Wikipedia has released the best and most practical breakdown of this topic, including an explanation of why LLM write the way they do.

A text can be AI-generated if:

➡️ The tone is not completely neutral. LLMs emphasize the significance of a topic or issue, using phrases like "marks a pivotal moment" or "though it saw only limited application, it contributes to the broader history."

➡️ Vague references to authority and evasive language. Examples include phrases like "observers have cited..." or "some critics argue..."

➡️ A superficial analysis. Even if a chatbot quotes a source, it does not necessarily guarantee that it has analyzed it in depth.

➡️ Parallel constructions. For example: "not only…but…" or "it's not just about…, it's…"

❗️ Wikipedia editors advise against relying on AI-text detecting services.

"While they perform better than might be achieved by chance, these tools have non-trivial error rates and cannot replace human judgment," the article explains.

Overall, if a text reads like an ad, it's most likely AI. This is because the internet is flooded with marketing copy, which AI is trained on.

💡 Of course, using em dashes (now a meme) or favorite AI stylistic patterns doesn't automatically mean the author used a chatbot. The main criteria are common sense and attentive reading. AI tends to produce a recognizable mix of templates, generalizations, and "significance." The better you understand these patterns, the easier it is to distinguish them from a real author's voice.

@hiaimediaen