[refulgentis]

> "This was AI" itself has all the tells of an irrational panic which typically accompanies new technology,

Being able to tell who wrote something doesn't imply irrationality, panicking, or a reaction to new technology.

> like UFO sightings in the 1950s.

UFO sightings stayed confined to the 1950s and were a reaction to new technology?

Or were the UFO sightings in the 1950s the only UFO sightings that were a reaction to new technology?

I'm not sure how this being clarified will be able to explain how identifying the writer of text is the same as a UFO sighting in 1950, but I'm open to it, I try to stay rigorously rational (c.f. X does not imply Y in first pull quote)

> If ever it is still possible to "tell" AI writing, it soon will not be.

Why not?

n.b. I quit my job at Google to build an AI client and have been working on it full time for 3 years. I love AI. I don't think there's a rational argument that justifies the idea it's better to never opine the author of some writing was AI, and the arguments offered here are particularly weak, at their face. As an opinion, solely? Fair enough.

[bluebarbet]

So I just read the article a bit more closely, and personally I see no reason to panic like you (and others here) are doing. The AI suspicion was presumably triggered by one of the subheadings, which follows the "It's not X, it's Y" schema. At this point it's almost a meme that this betrays AI.

But I say: who cares? The substance and the authenticity are what count. This article made some interesting points, and it was signed off on by a human author. Personally, I'm no more interested in whether the author used AI to produce the text than in whether they used a dictionary or thesaurus, as long as they stand by the words.

This whole "debate" has the feel of religion to it. I'm consistently surprised that there's so much woolly, unfalsifiable thinking on this subject. And here, of all places.

[drdexebtjl]

It’s not just that subheading, it’s a recurring pattern.

It’s not about authors using AI, it’s about authors not respecting the reader’s time.

Writing has a multiplicative force. The writer only needs to spend time writing it once. The text will reach thousands of readers who will collectively spend a lot more time on it than the author.

If even with that asymmetry, the author is not willing to put effort into writing it, I’m not going to bother reading it, regardless of the text’s substance. I’m just not interested.

[ranger_danger]

I think the main problem is you can't really ever tell with a high degree of certainty, people are just guessing based on what they see in an unscientific way. And the fact that AI is trained on human data, meaning what we see is in fact things humans have already done themselves, makes it even harder to "know" for certain IMO.

[refulgentis]

> the main problem is you can't really ever tell with a high degree of certainty

This is false, it is trivial to find humans with 99%+ accuracy[1] and there is a well-known service with 99%+ accuracy when analyzed by 3rd parties with no affiliation.[2]

> people are just guessing based on what they see in an unscientific way

As we see above, this is just guessing in an unscientific way. :) It's important to be rational! We all agree on that. :)

FWIW it did used to be true that the 3rd party services were junk. And I don't support the idea of humans just winging it when there's consequences. The case we're in is about the most mundane and consequence-free, the vast majority of us use AI daily and we're commenting on an internet aggregator that is aggregating a blog article.

[1] People who frequently use ChatGPT for writing tasks are accurate and robust detectors of AI-generated text - https://aclanthology.org/2025.acl-long.267/

[2] Artificial Writing and Automated Detection” - https://bfi.uchicago.edu/insights/artificial-writing-and-aut...

[ranger_danger]

> This is false, it is trivial to find humans with 99%+ accuracy

Sorry but no, this is also false. AI is largely trained on human data. What it outputs is also largely what a human would output. We could both easily make up the exact same sentences (especially smaller ones) and there is NO way to tell what the real source of a sentence is, especially considering the source can be multiple things (AI and human) at the same time!

There is no amount of literature you could provide or anything you can do or say that would change my mind on this.

[drdexebtjl]

Humans can write these tropes individually. But we usually don’t write them one after the other over the course of the entire text.

We have our own personal biases towards certain words and sentence structures that don’t match the same biases that post-training favors. We speak egoistically how we want to speak, not necessarily how the listener wants to hear.

We don’t polish every sentence, because not every sentence needs polish.

If a human wrote like this, I would be concerned that they’re spending too much effort into making me feel or thing something, and not enough effort into expressing their own thoughts and emotions. I would be concerned that they’re spending way too much effort into writing blog posts, and not enough effort thinking about the topic they’re writing about. I wouldn’t want to read that either.