Some Interesting Things I Read Today (12 May)

I want to stop linking to archive.today because of its troubling behaviour, so I won't be providing that link for some sources. I also won't link directly to 404media because they annoy me with their “it's free, but give us your email” tactic. You'd think an outlet writing on privacy would realise how gross that is, but...

I also don't link to Substack because they have knowingly and directly supported Nazis; I wish people would move off and to something like Ghost, but I can't also pretend to know their reasons for staying on the Nazi Blog Platform. I don't think this people are bad, btw; I just think they have another priority, and I don't know what it is. Anyway, I tried archiving via IA, but Wayback Machine throws out error pages that make you look for the relevant post and then click a link in the suggestions... which makes zero sense.

I will be going about my day reading, watching, or listening to something and, suddenly, I notice that something is wildly off. Quite simply, I feel like I’m going nuts.

I feel this very strongly. I don't normally assume everything is AI, but when it starts hitting me in ways that feel strongly of AI... It makes me feel like I've lost the plot. I also feel very strongly that, even when people tell me they've used the generative AI in certain parts of the development of something... That I find it hard not to question whether they've been using it elsewhere, and it makes me want to not engage with anything else they do.

(Disclosure: Pangram Labs previously advertised on 404 Media).

This is a fun thing to point out because AI detection is shit at detecting AI... and often flags non-AI things as AI because (in a fun ouroboros of shit, AI is trained on non-AI things in order to create the “more human”-sounding AI bullshit). So while I'm glad 404 did disclose this in the article, they also need to not promote people who promote the use of AI in ways where it doesn't quite work.

Other fun things to note:

When I sat down to write this article, in which, to be clear, I did not use AI, I found myself writing the following sentence: “It’s not just in places we’re conditioned to see AI—Google AI overviews, LinkedIn influencer posts, and Facebook feeds—I’ve started seeing AI…” I stopped typing, freaked out, and deleted the sentence. Have I always written this way? I honestly don’t know.

This AI construction is a common way of speaking. Of course you've likely written like this, but it's the frequency. How often do you say “It's not x, it's y?” That's the bigger tell for AI rather than whether a person uses that phrasing or not. Same thing goes for other so-called “tells”: the em dash, semi-colons, and parentheticals. I use a lot of all these things and have for years, but what is the comparison between how much I use them and how much AI uses them? (And in some cases, my use of some things mirrors AI despite the fact that I know I've used them “excessively” and can go back to examples from the 2010s to prove it.)

Last month, the New York Times quoted a teacher who said “They are using generative A.I. to write before they learn how to write.” Teachers I spoke to last year lamented that they are spending their very real human hours and considerable brain power trying to determine whether they are grading essays that are written by humans or robots, and know that they are often giving writing notes on papers that were likely written by AI.

Speaks more to the school system. When kids don't feel they're doing something within their capabilities (like kids who struggle to write in the language of instruction because they're not getting necessary language support) or something that is authentically engaging, you're going to end up with AI work (much like you end up with copied work prior to AI). Also, it's easy to tell whether the work your students give you is AI if you work with them throughout the whole process and also engage them in other things (e.g., note-taking for the process). I know a teacher's job requires a lot (I've done it for 15 years), but I'm seeing a lot of focus on the wrong things and teachers handing out excessive amounts of work.

This was linked in the above article, but when I opened it and saw:

The replies pointed out something crucial, something that makes this whole debate even more infuriating: Some of us actually had to learn English.

I knew I had to actually read it. Even though I grew up speaking English as my first language (and therefore didn't have to learn English in the same way because it came to me “naturally” as part of the environment I grew up in), I have worked with a number of people who speak a different first language and therefore learned English in a formal classroom setting.

To these detectives of digital inauthenticity, I say: Friend, welcome to a typical Tuesday in a Kenyan classroom, boardroom, or intra-office Teams chat. The very things you identify as the fingerprints of the machine are, in fact, the fossil records of our education.

And even though the people I have worked with are Korean, Slovak, Czech, Polish, Chinese, Ghanaian, etc... They also frequently use the same “AI tells” that Marcus lists because they all learned English in a classroom with varying degrees of strict colonial attitudes built into the curriculum.

Recent academic studies have confirmed this, finding that these tools are not only unreliable but are significantly more likely to flag text written by non-native English speakers as AI-generated. (And, again, we’re going to get back to this.) The irony is maddening: You spend a lifetime mastering a language, adhering to its formal rules with greater diligence than most native speakers, and for this, a machine built an ocean away calls you a fake.

I just love this paragraph, so go read Marcus's piece. I'm not saying more than that.