Ivy: Eight-hundred and sixteen billion parameters... on a desktop computer.
Marcus: I'm Marcus.
Ivy: I'm Ivy.
Marcus: And this is Broke It for Monday, October 19th, 2026. Here's what we're breaking down for you today in AI.
Marcus: Meta and Apple claim they've crammed a monster AI model onto a Mac Studio. We'll see if it actually works.
Ivy: Then, Meta quietly deleted a single word from its terms of service… a word that says everything about the future of your data.
Marcus: And Disney is making a huge play for the team at Character.AI, but the deal is a lot messier than it looks.
Marcus: Alright Ivy, let's start with squeezing an elephant into a shoebox: Meta's new Muse Spark Open model... on an M5 Mac Studio.
Ivy: An 816 billion parameter model, to be clear. That's an astonishing number. Models that big usually need entire server farms.
Marcus: And they're running it locally on Apple's MLX framework. This is the dream, isn't it? Powerful AI running right on your desk, no cloud required.
Ivy: It's a dream that comes at a cost. The only way they can pull this off is through extreme quantization.
Marcus: Right, so quantization, for anyone who doesn't know, is basically shrinking the model by using less precise numbers. Think of it like using fewer decimal places to make it smaller and faster.
Ivy: It also makes it dumber. The report cites 'significant performance degradation' and says it pushes even a top-tier M5 Studio to its absolute limit.
Marcus: But even a degraded 816-billion-parameter model is still incredibly powerful, right? It's a proof of concept! It shows the hardware is finally catching up.
Ivy: It shows you can do it, not that you should. It's a technical flex. But is it a useful product? The jury is way, way out.
Marcus: So what's the takeaway for the average person?
Ivy: It means the bleeding edge of local AI is going to be messy. You might get that flashy model on your machine, but it’ll be a compromised, power-hungry version of its parent in the cloud. At least for now.
Ivy: Sticking with Meta... they just made a tiny edit to their data policy that has some... massive implications.
Marcus: Okay, I saw this. They changed their terms for crawling public websites, right? What's the big deal?
Ivy: The big deal is one word they removed: 'train'. The old policy said they could use public info to 'develop and improve services, including to train its models'.
Marcus: And the new one?
Ivy: The new one just says for 'providing and improving our services'. The specific mention of training AI models? Gone.
Marcus: Wait—so they're not training on public data anymore? That seems... unlikely.
Ivy: Of course they still are. They just swapped a specific, controversial term for a vague, all-encompassing one. 'Improving our services' can mean anything... including training their models.
Marcus: So it's a legal move. They're trying to sidestep lawsuits from creators and publishers who don't want their content used to train Llama 5 or whatever comes next.
Ivy: It's preemptive legalese. By removing the word 'train', their lawyers can argue in court that scraping your blog to make their image generator better is just part of 'improving the service'. It's a semantic shield. And honestly? It's kind of slimy.
Marcus: So for the rest of us, what's the bottom line?
Ivy: It means the fight over your data isn't happening in big public announcements. It's happening in quiet revisions to the fine print. They are betting you won't notice them broadening their rights to your digital life.
Marcus: Alright, last up today… from data rights to character rights. Disney is making a move on Character.AI.
Ivy: Well, let's be precise. It's not a full acquisition. A new filing shows Disney has agreed to hire most of Character.AI's team.
Marcus: An acqui-hire. But still! The minds behind the best conversational AI characters joining the House of Mouse? Imagine talking to C-3PO, or Elsa, or Buzz Lightyear. This is a perfect match.
Ivy: Hold on. The filing also clarifies this is a 'stalking horse bid'.
Marcus: A what now?
Ivy: A 'stalking horse bid.' Basically, it's an opening offer on a distressed company that sets the minimum price. So Disney made the first move, but now other companies can jump in with a better offer.
Marcus: So... Google or Microsoft could swoop in and snipe the whole team away from Disney?
Ivy: Potentially. Disney's offer just sets the floor to start a bidding war. The good news for them is that the vast majority of employees offered jobs have already accepted. It's a strong signal, but it's not a done deal yet.
Marcus: Wow. Okay, so what's the big picture here for interactive entertainment?
Ivy: It means the talent that can build believable AI characters is now one of the hottest commodities in tech and entertainment. This is the starting gun for a bidding war over the people who will define the next generation of interactive storytelling.
Marcus: Alright, let's recap the day.
Ivy: Meta's showing you can run a massive 816-billion-parameter model on a Mac, but it's not without some major compromises.
Marcus: And Meta is playing word games with its data policy, trying to quietly grab more rights to train on your content.
Ivy: And Disney has fired the first shot in a potential bidding war for the talent behind Character.AI.
Marcus: Before we go, Ivy, if this Disney deal goes through, what's the first character you're talking to?
Ivy: I suppose I'd ask Scrooge McDuck for financial advice. At least that has some practical application.
Marcus: That's our show! I'm Marcus.
Ivy: And I'm Ivy. Thanks for listening to Broke It. We'll be back tomorrow, hopefully with some hot investment tips from a cartoon duck.
This show is made with AI: the hosts’ voices are synthetic and the scripts are AI-assisted. Every story links to its original source.