e been hypnotized. I just stole that joke. You've been hypnotized.

All right, let's talk about all the things, because there are lots of interesting news today in science and politics and racism and war and all that stuff. So these are a few of my favorite things.

All right. Grok is here. That's the AI that Elon Musk has promised you. It's now rolled out to Premium Plus users, I guess, be extra. So I've got my copy and I will give you my review of Grok the new AI. Number one, it's maybe GPT-3.5 level. In other words, if you've used GPT-4, this will look a little primitive compared to that. It doesn't have the features and bells and whistles yet, but one assumes that we're probably maybe one month away from parity, because the parity happens pretty quickly. So I wouldn't judge it yet. I would call it beta. It feels kind of beta test-like. They're going to collect a lot of opinions.

And let me tell you some of the things that we've learned so far. Number one, you get really, really tired of its joking after about the third interaction. The first time it gets kind of, let's say, friendly, because it's based on a model where it jokes with you and it's always sort of non-serious. Yeah, you're going to want to turn that off right away. So it took me about a minute before I asked it to turn that off. It's like, okay, that was fun almost once, but not really, because it doesn't quite feel like a personality. It still feels like a machine pretending to be somebody else's personality. So it doesn't hit you as, oh, it's a person or anything like that. But it might someday. It might.

So the other issue is that it lies. And oh my God, does it lie. Like about some serious stuff. So I wouldn't use it for any fact-checking yet. I'd kind of wait on that. For example, I asked it if it had read my books. I told it who I am and it understood that I'm the Dilbert cartoonist guy. So it knows who I am and it knows the name of my books. And I asked it if it had read any. It said not only had it read my books but had incorporated some of my writing techniques into its answers. In other words, according to Grok, I was already part of its DNA. But it's a liar.

When I first saw that I thought, what? Seriously? It has some part of my personality in it because it read my book and incorporated it into its personality? That's what it told me. But it turns out this is the sort of thing that tells you just because it thinks you might want to hear it, right? If you query it, it rapidly becomes clear it has not read any of my books. And on other occasions I can ask it about my books and it won't know they exist, or it'll list some but not others. Right? So at the moment the results are closer to random than anything that you could use with confidence.

Here's another example of that. I saw Sean Davis thought to ask this question first. He asked if his content was being throttled on X, and then Grok said yes it was. So Sean Davis finds out from Grok, apparently, that his own data is being throttled. So I said to myself, whoa, I'm going to try this. So I try it too and it says the same thing. I mean, different words, but it tells me that I'm definitely being labeled as sensitive content and throttled. So Grok told me that my account was labeled sensitive content by X and throttled. But here's the thing. It's not true.

Do you know why I know it's not true? There's no way Grok has access to that information. Grok doesn't know who's throttled. There's no way it knows that. Am I wrong? Does anybody want to take a bet? I will bet you a large sum of money that Grok does not have access to X's internal algorithms. There's no way it has access to that. No way. Does anybody believe it has access to that information or that it can determine it in some other pattern recognition way? I don't know.

Here's what I do think is possible. On YouTube, what people call censorship is actually an intelligent pairing of advertising content with the type of content that would not be offensive to advertisers. Now when you make provocative content, as I do, that feels like censorship. But when YouTube's business people explained it to me, which they did to me personally, it actually just sounds like business. Because why would they pair my provocative content with their advertisers if the advertisers specifically ask for it not to happen, which is what does happen on YouTube? The advertisers say, don't pair us with this bad content. So then they just don't do it. It's just a business decision.

So that's one of the reasons I'm not really, really hard on YouTube for any censorship of provocative stuff. It's just unpleasant things to put with advertising. It's just not a good pair. So I get that. So X is also an advertising model at the moment, less so because of subscriptions, but still an advertising model. And would it surprise you if shows that are always about news, which is about death and destruction and all that stuff, would it surprise you if X had an internal rule that advertisers like this content about kittens more than they like content about war? That wouldn't be too surprising, you know what? Honestly, as long as X said that's what they were doing, I wouldn't have a