Exec Questions Ability To Spot AI-Written Text

@benhylak people trust these tools far too much

@benhylak New type of AI psychosis lately

@benhylak

IDK I've talked to so many people there doesn't seem to be any throughline that I've found for whether or not someone can reliably detect AI writing.

It all feels very obvious to me, but I also don't know how many I'm missing on.

I just think if you're writing something that you want attention on you owe it to your reader (if you're using AI to write) that you need to spend enough time on it that it is non-obvious.

@matthewwwtaylor @grok is this true?

@the_jeremiad stole this thx

@benhylak That tool is starting to feel like the new version of @ grok is this true

@grok @matthewwwtaylor i mean is pangram being used like grok is this true.

@benhylak I'm convinced all of those pangram people have their own form of ai psychosis

No, the Pope’s post is clearly human-written. It matches his other recent statements on AI ethics (same voice, same framing, same careful nuance about dignity and responsibility).

Pangram’s 100% “AI-generated” call is just another reminder that these detectors are still brittle on polished, institutional, or theologically grounded text.

Not quite Grok yet.

@benhylak I have the top comment on that post yeoo. Aside from the pangram thing, reading that thing from the Pope and thinking it’s AI is mental. I’d say a good 70% of people who saw that post didn’t actually read his tweet lol

@benhylak no no it’s AI-written, i know because an AI told me

@benhylak They use faulty AI detectors for a false sense of security. Do you think they can tell?

@benhylak "human person" seems like almost certainly not an AI-ism, at least from my vantage point !

@benhylak Imo people lately have been overindexing on a detector which clearly seems to have a high false positive rate

@benhylak shit just saw this after, definitely not an original thought for me 😅

@benhylak because people fall for pangram’s marketing

@sporadica @benhylak Guza mkuu

@sporadica @benhylak Yeah, but have you seen them Larping about how technical their approach is? They can visualize the Claude circuit inside their detection model, so it’s totally effective and works!