Brian Barrett: And also the kind of thing that people readily click through and don’t realize is happening, right? TikTok also now tracking data. Anything you put into any of its AI tools, if you put in a prompt or any kind of info you give it there, it’s going to track that data, it’s going to use it.
Again, all of this is ultimately used to serve you ads, but it is the principle of it, and it is the sort of lack of the opacity of it, the lack of clarity, just sort of, oh, all of a sudden this is happening. And I think it is a legitimate thing to keep a close eye on, even if it hasn’t happened yet, how TikTok’s algorithm changes in more subtle ways.
I don’t think if there is going to be some kind of manipulation of TikTok, I don’t think it would be as ham-fisted as “You can’t post about ICE.” I do think you can see things changing in terms of what people encounter in their feeds and what gets promoted and what’s not, because that fits in the black box. That’s something that you can’t quantify from the outside, but can sort of have an effect on user sentiment, which is, by the way, the exact thing they were trying to get around happening when they started this process in the first place, of getting TikTok out of Chinese ownership.
Tim Marchman: Another problem here is just that this is with TikTok in particular, an infamously difficult thing to track. A few years ago, when researchers had API level access to X, for instance, they were able to do a lot of really impressive sentiment analysis and just tracking how things worked. Without that kind of access to X now, that work is difficult to impossible. And TikTok is much slipperier because it’s such a personalized algorithm. So one thing to just be aware of is that we may not be able to track, with a lot of confidence, changes made to what it’s serving users, even if we’re anecdotally pretty sure it is changing. So it’s not a fun thing to be thinking about.
Zoë Schiffer: Now that we’re talking about apps, I’m going to pull us to our lightest segment yet.
Brian Barrett: I like how we’re progressing.
Zoë Schiffer: Yes.
Brian Barrett: We’re just sort of easing up as we go.
Zoë Schiffer: Exactly. I think this one’s genuinely fun.
Tim Marchman: We’ll leave here in a great mood.
Zoë Schiffer: Yes. Have you guys heard of ClawdBot, now known as MoltBot?
Brian Barrett: So I have, but I desperately need you to explain why I should or should not buy a Mac Mini, and put it on it and use it.
Tim Marchman: Same. This is in my peripheral vision, but I’ll be learning as we go here.
Zoë Schiffer: I think Tim will need to explain the immense security risks associated with letting this app run your life. But basically the magic of this app, as I understand it through Will Knight’s great reporting on WIRED.com, is that it’s basically an AI assistant that connects a lot of different apps on your computer. It runs locally, and crucially, you can talk with it over a messaging app. So send it commands, and then it’ll go figure out how to execute and run your life for you. The caveat, as with all AI assistants, is that when you hear, “Wow, AI assistant is going to run your life.” At least to me, I’m like, “Can it submit an expense report? Because I would give it any amount of information to make that happen.”
