My day job is a copy editor at a daily newspaper. Although I spot signs of AI writing, it’s quite harder to be doubly sure. That I even find the difference at all is thanks to the way Indians write in English Vs Americans or Brits. For instance, we don’t say “individuals” often. We prefer the word “person” or “people”. Since ChatGPT uses “individuals” a lot, I know that a colleague has actually used the AI tool when I spot “individual” in their article. But then, I also know that if you keep reproducing the content with prompt for changes, you can completely camouflage AI writing. Also, it’s quite hard to dispute if someone says that the original idea behind an article is their own and they have only taken the help of AI tools to express it correctly.
I've posted similar questions to vendors who claim to have similar results in identifying AI-generated resumes and cover letters. Again no responses. Complete snake oil in my opinion.
True, false positives can be the most frustrating part because they waste time and make people lose trust in the whole system. I keep a quick note in the middle here [https://lifecircu.com/types-of-listening/](types-of-listening) on why clear feedback loops matter, and in practice the best fixes are better thresholds, better training data, and an easy way for users to report “this was flagged incorrectly” so the model improves over time.
This post provides a clear, well-researched, and balanced perspective on AI detection tools. By using OpenAI’s own data and raising unanswered questions about bold accuracy claims, it highlights the real risks of false positives and the need for transparency especially when the consequences for users can be serious.
Thanks. I am generally neutral about that topic. In some cases, it will be deemed important but in most I don't think it will. Some journalists have the luxury of review and polish by editors. For other journalists, ChatGPT may provide that service. I am particularly interested in how these technologies can help non-native speakers of the language they are writing in.
As you have probably guessed, I'd like to see disclosure of the data that supports claims that tools can identify AI written text. There are existing protections in many countries against making false claims. Is there an ethical requirement to proactively disclose evidence of efficacy.
My day job is a copy editor at a daily newspaper. Although I spot signs of AI writing, it’s quite harder to be doubly sure. That I even find the difference at all is thanks to the way Indians write in English Vs Americans or Brits. For instance, we don’t say “individuals” often. We prefer the word “person” or “people”. Since ChatGPT uses “individuals” a lot, I know that a colleague has actually used the AI tool when I spot “individual” in their article. But then, I also know that if you keep reproducing the content with prompt for changes, you can completely camouflage AI writing. Also, it’s quite hard to dispute if someone says that the original idea behind an article is their own and they have only taken the help of AI tools to express it correctly.
Great context. Thanks for sharing.
I've posted similar questions to vendors who claim to have similar results in identifying AI-generated resumes and cover letters. Again no responses. Complete snake oil in my opinion.
I agree.
True, false positives can be the most frustrating part because they waste time and make people lose trust in the whole system. I keep a quick note in the middle here [https://lifecircu.com/types-of-listening/](types-of-listening) on why clear feedback loops matter, and in practice the best fixes are better thresholds, better training data, and an easy way for users to report “this was flagged incorrectly” so the model improves over time.
This post provides a clear, well-researched, and balanced perspective on AI detection tools. By using OpenAI’s own data and raising unanswered questions about bold accuracy claims, it highlights the real risks of false positives and the need for transparency especially when the consequences for users can be serious.
Also, am going to cite this post in my piece on the ethics of disclosure or non-disclosure of using AI tools in writing.
Thanks. I am generally neutral about that topic. In some cases, it will be deemed important but in most I don't think it will. Some journalists have the luxury of review and polish by editors. For other journalists, ChatGPT may provide that service. I am particularly interested in how these technologies can help non-native speakers of the language they are writing in.
As you have probably guessed, I'd like to see disclosure of the data that supports claims that tools can identify AI written text. There are existing protections in many countries against making false claims. Is there an ethical requirement to proactively disclose evidence of efficacy.