AI

VocabularAI

Posted on

    GPTZero lists words and phrases that are common indicators of #AI. The problem is that the words and phrases themselves are very common: “consequences”, “such as”, “primarily”.

    These supposed AI-detection tools always make me uneasy. It’s hardly surprising that AI has a strong predilection for cliches of formal #writing; that’s what it’s trained on. It wouldn’t surprise me if #Wikipedia content alone has a noticeable impact on how these models write.

    But the upshot is that it’s getting harder and harder to not sound like AI. I try to avoid cliches to begin with, but if I’m writing something boring and technical (like for work), that’s not always easy.

    I understand the suspicion around AI. What’s worrying—and arguably as socially/culturally corrosive as the technology itself—is the superstition around it: seeing AI in every shadow, whether it’s there or not.

    Posted on

      The #Apple logo banner on apple.com has a distinctly Siri-esque shine to it today.

      This being a Monday, I don’t have time to watch #WWDC, but I’ll be very curious to see if whatever Apple rolls out here fixes anything. #Siri is probably the most conspicuously underpowered product in Apple’s lineup right now, at a time when Apple is—supposedly—working on small wearable devices that do not have touchscreens and will, I assume, use Siri as their primary interface.

      I have one such product already: a HomePod that I never use, mainly because I hate dealing with Siri. The fact that Apple is apparently upgrading Siri with a custom version of Gemini isn’t encouraging; we use #Gemini at work, and I haven’t found it to be nearly as capable as either #ChatGPT or #Claude. Still, Apple is not in a great place AI-wise right now, and the new #AI guy, Amar Subramanya, ran Gemini at Google, so I guess Gemini is the most obvious option.

      Best case, Siri turns out to be like Apple Maps: a rough start, but improving year over year.

      Worst case, Siri turns out to be like Apple Maps: even after 14 years of improvements, it still hasn’t fully caught up.

      Say please

      Posted on

        I’m starting to think that the best way to conceptualize #AI isn’t as vast, potentially malicious, hyper-advanced systems, but as vast, potentially malicious, hyper-advanced systems that power the most gullible, bumbling assistant you’ve ever worked with.

        Like the sort of eager-to-please person who would give hackers access because they asked nicely. Or the overzealous go-getter who deletes all your emails and then apologizes for it.

        Both at #Meta, by the way. One gets the sense that that “move fast and break things” cultspeak has not scaled so well.

        Your favorite movie is vertical now

        Posted on

        A friend recently shared this video from Kendra Gaylord, describing how your favorite movie is vertical now, courtesy of #AI —and what is lost in the process:

        In movies, every time something is filmed, it’s memorializing a lot: the actors, the location, the way we talk and write and joke. Even though it’s fiction, it’s still a document, and if you mess with that document too much, it doesn’t represent all of those things anymore.

        One of the most perverse revelations here is that the AI-generated version felt more realistic to people simply by virtue of being vertical; the same aspect ratio in which we experience social media videos. Our window to reality is portrait-shaped.