28 Comments
User's avatar
Ellen Batchelor's avatar

I'm so sick of reading ai generated articles. Same tone. Same style overall. I can't read more than a couple paragraphs and unsubscribed from more than one content maker who used to be good as a result. I'd rather have fewer authentically written pieces than lots of drivel.

CrossTieWalker's avatar

I’ve had the experience of being accused of pasting AI content even though it was all my own. I am an old-timer who believes in correct spelling, punctuation, and proofreading, as well as some attention to composition and diction.

The letter M's avatar

I just wonder how many human articles will be flagged as false positives?

Substack Reader's avatar

"If it floats, it's AI!"

Ellen Evans's avatar

If the article weighs the same as a duck, it's AI.

Tardigrade's avatar

That's a real concern.

Gargoyle Protocol's avatar

Lawyers, accountants and doctors are also trying to get keep their intellectual advantage so it's only natural for the word Smith's to attempt to gatekeep along with the intellectually elite academic class. Unfortunately time is not on your side. It is the formation of the human mind that is at risk and that is what should be protected. Not the tools that one employs but rather how one's thoughts are formulated. These are different conversations.

Gargoyle Protocol's avatar

Outsourcing individual preference? And discretion to a third party is never useful. The argument is that you're going to use an AI to tell you whether or not the thing you're reading is from an AI. You're going to outsource your own discretion to the machine you're afraid of. And for the authors who's livelihood depends upon revenues from their subscribers, subscribers who might be turned off by a false positive this could be devastating. On the one hand, it's censorship and gatekeeping and on the other hand, you're just outsourcing more of your intellect to a machine... This subject is exactly the ontology and teleology of my substack and this is why I will never charge for any word posted.

David Parker's avatar

Freddie DeBoer has a good article about this. He found the program to be inaccurate. A 300 word excerpt from one of his posts was judged to be AI written but when the whole post was put in it was judged to be 100% human written. He got other inconsistent results with it. I would be highly skeptical of it.

JollyLittlePerson's avatar

Couldn't this be used as another means of control? Any article or comment that says something they don't like, is flagged as AI content? Someone is setting up the algorithm to find something - so what if it's flagging dissent and unacceptable thoughts rather than AI writing style?

Efferous's avatar

FWIW, I tried out the underlying tool and found it very inaccurate. A 3000 word test article I fed into it came back with the claim of 100% AI-written. I suspect there will be a bit of a witch hunt coming.

Substack Reader's avatar

A new Turing test... Can an AI write something that another AI will think was written by a human?

Ian Landesman's avatar

Matt, I would rather infrequent actual words from you than frequent AI slop, even if curated to be as good as possible. You have my subscription $ because you tell the truth sporadically in a sea of BS. Keep doing what you are doing!

Tardigrade's avatar

There are one or two Substack authors that I would look forward to running through this analysis >cough< Malone >cough<, based largely on a sudden avalanche of output.

However, I really hope this analyzing machine isn't heavily weighted against em dashes. We em dash aficionados are feeling unfairly persecuted. 😜

Substack Reader's avatar

Anyone know if I can run this on my own comments? I would like to see how I stack up. I don't know if it would be a compliment or an insult to rank highly on the AI meter.

Hollis Brown's avatar

looks like I’ll be keeping my comments to under 100 words to retain my mystique.

958189's avatar

The tool can be helpful when the content discussed is technical or mechanical eg., off grid electrical systems, especially for technical information that is not well understood or available locally. That could be an important distinction between AI models. What actually works safely in the technical sense is separate from opinion or choice based on ideology.

CE Shady's avatar

I suppose that could be useful for pointing out bots in the comments section (if it works) but is there anyone who seriously can't broadly spot AI generated content?

It's so easily clock-able, in both email solicitations and articles. LLMs have "a voice", and it's a pretty obvious one. The waste is in needing to click on the article, or read the email, and realize "oh, this is AI" before ignoring it and going back to your day. It would be sad to need to "run it through a tool" to save yourself from content creators you already pay a sub to.

Lekimball's avatar

This is great, Matt. I am having company for dinner, so I hope you post a transcript.

Substack Reader's avatar

Cancel the dinner, for crying out loud.

Lekimball's avatar

Very funny, but no.

Katie Andraski's avatar

Seems to me this is a witch hunt. I think there will be false accusations. I barely remember to put a share a comment box let alone my post is completely human made. Are we not allowed to use AI as a search engine? Thank you for doing this.

Taras's avatar

AI-generated content should be labeled as such, like any other source you quote.

Tardigrade's avatar

AI is very useful as a search engine.

I think the objection is to authors who flood us with content that they haven't really written themselves. As with everything, it's a spectrum, and the question is where do you draw the line?

Katie Andraski's avatar

There’s also the issue of our content being scraped to educate AI. This is all feeding the machine. My content is not AI generated.