AI生成内容的蛛絲馬跡 Tips for spotting AI-generated content
Asked AI to output an article on “How to detect AI-generated content” in less than 400 words and it returned the following:
"With the rapid advancement of artificial intelligence (AI), detecting AI-generated content has become increasingly important. Here are some key strategies to help you identify AI-written text..."
Substance-wise, such an introduction is empty because it is there for the sake of being there without insight. It is also not communicative. But don’t blame it. AI content creators are programmed to always provide articles with an unopinionated introduction.
Enough for the human-written lead-in you are reading. Let’s get to the point:
Tip 1: The two red flags of "remember" and "in today’s world"
AI just tends to use the word "remember" to start a sentence to reinforce some idea, such as:
"Remember, the key to a great outfit is not just the clothes but also how they fit and how confidently you wear them." (In fact, “not only…but also” is also favored by AI.)
Next, it is "in today’s world":
"In today's world, cryptocurrencies have emerged as a disruptive force, challenging traditional financial systems and offering new possibilities for decentralized transactions and digital assets."
Abandon them, even for humans because they are cliché of cliché.
Tip 2: Absolute adherence to grammatical rules
Humans dare to and know how to break the rules by using unconventional grammar and phrasing to express humor, emotion or originality. In contrast, AI prioritizes getting things right.
Tip 3: Directionless and detached responses
Avoiding biases, AI does not express strong opinions, clear positions or self-reflection. AI-written content sounds cautious and detached without direction or personal content.
Tip 4: Details and facts
AI-generated articles could contain factual inaccuracies.
Tip 5: Real-timeness of information
AI language models are trained on data up to a certain point in time, for example, 2021 or earlier. Content discussing current affairs is one sign of human authorship.
Tip 6: The writer's background
Look into the author's background and previous work. Mismatch between past work and the current article's content could indicate AI involvement.
Tip 7: Catching it red-handed
Phrases such as "regenerate response" and "as an AI language model, I" indicate direct copy-and-paste from an AI chat window.
Tip 8: Rare use of brackets and em dashes
AI likes straightforward sentence structures, for example:
Human writer: Mary always brings her dog (the one that barks at me) when she comes to visit.
AI writer: Mary always brings her barking dog when she comes to visit.
Tip 9: Using AI content detectors
AI content detectors are there to help identify AI-generated text but beware those tools are not always accurate. Here is one with free trial as of the date of this writing:
It will be increasingly difficult to detect AI-generated content as the technology improves. The "This This Rice" strategy (a combined mode; here it refers to "technology + own due diligence skills") may be more viable.
Commenti