Is Meta Done Sharing Their AI?

**Brian** (0:00)
Hey, what's going on, everybody? Welcome to The Daily AI Show. Today is July 30th, which as I learned this morning from White, is not the last day in July, but it's July 30th, it's Thursday.
We're live, it's 10 a.m. on the East Coast, it's 7 a.m. on the West, and we were just all being old people right before the show started, and all talking about foresleep and saying how it's a little harder if you're on the West Coast to be up and ready and prepared for the show. And I was saying, I don't know that I could do the show at 7 a.m. So kudos, Gareth and Andy and Jyunmi and you and Karl, to a certain extent, Beth and I have a little bit easier on the East Coast. Although, Beth, you are sort of admittedly a night owl, aren't you? Yeah. Yeah. So maybe 10 a.m. is early in the morning as it may seem for others.

**Beth** (0:47)
I feel like you can tell when on the show, I have not made words out loud before the show starts.

**Brian** (0:53)
Yeah.

**Beth** (0:54)
Like, oh, hello, morning, boys.

**Brian** (0:56)
It's so funny you say that, Beth. Amanda, my wife, Amanda, literally came into the bathroom this morning as we were trading places. We get up about the same time, about 6.30, and she went to say, like, hi, babe, and it came out like, hi, babe, and she's like, oh, I'm the first person you said anything to this morning, aren't I? She's like, yeah, that's weird. It's always funny when you get out on a client and you're like, sorry, sorry, you're the first person I've talked to today. So anyway, so that's what's going on in our neck of the woods. But we wanted to get in obviously into the AI stuff. Hey to Jeff Hardy in the comments, appreciate you being here. And yes, Gareth is sans mustachio today. So always a new look, always a little something different, Gareth is keeping us on toes.
Andy, I know you were saying, because I was asking you, you're like, yeah, busy morning and I was like, oh, on the farm. You're like, no, no, no, in AI.
So why don't you kick us off with what you were talking about? Because we didn't get to that.

**Andy** (1:53)
Yeah. So the first thing I woke up to was that Microsoft's beat of Wall Street expectations yesterday, pumped their stock up a little bit. Meanwhile, Meta reporting their ballooning costs, they went down. And so that opens, the reason for Microsoft's beat was largely the revenue from Azure Cloud, which is where their AI business sits in their pantheon. They also are developing their own models to run on Amazon. I'm sorry, not Amazon, Azure Cloud. And so anyway, that was an interesting perspective that Microsoft's doing very well with their AI business and they're reducing their own internal costs of supplying it by using cheaper models than Anthropic and OpenAI models or making those available to their Azure customers. Then as regards Meta, Meta has a bunch of different ways of trying to address the news that their ballooning AI costs are crippling their economics. And so that they dropped 6% in after-hour trading. They posted a sharp earnings decline in the face of their very, very high investment in AI development, including infrastructure costs, etc. Whereas by contrast, Microsoft reported increased income, while also indicating through their reported financials that they're going to be investing tens and close to a hundred billion dollars of their money, their cash flow. And so neither of these is cash flow positive, but the reported earnings were good for Microsoft, bad for Meta. And then on top of that, a little point of interest here is that, while they've announced 8,000 new layoffs Meta has, they currently have 75,500 employees, which is only down 1% from the prior year. So despite these huge numbers that we think, okay, wow, there's 8,000 more people going to lose their jobs at Meta. That's a very small percentage of how big Meta is in headcount. So that's interesting. Okay, so then further on Meta.
Today, Zuckerberg published a policy paper in which he claims that Meta AI systems are now self-improving. So they have their own internal self-improvement loops and they're doing that.
He's not declaring this is the reason, but he's saying from this point forward, their most capable models are not going to be open release.
They're not going to be open source, because super intelligence labs is not going open source anymore. But they're not even going to be open release. They're going to be like Mythos and like the others who have secret very high-end models behind the scenes. They're not going to put those in front of the public. Why? Because the combination of self-improvement through recursive process, training itself to be better and better, and the advanced capabilities for hacking, those things combined to make it important for the major frontier model developers to keep those things under wraps until things can improve. To close out, the other thing that's wrapped into all of that that I was thinking about this morning, just on this one subject, which is what's happening with the frontier model developers.