Ivy: It’s not another text generator, it’s a decision generator. For your phone.
Marcus: [beat] I'm Marcus.
Ivy: And I'm Ivy.
Marcus: And you're listening to Broke It, your daily AI news for Sunday, October 11th, 2026.
Ivy: You bet. We're looking at a new kind of AI model—one built for taking action, not just generating text.
Marcus: Then there's Microsoft, promising to run a massive 120-billion-parameter model on a laptop... if you've got the cash.
Ivy: And Microsoft's own CEO says it's time for an 'emergency brake' on AI. The timing is... let's just say, interesting.
Marcus: So, let's start with Liquid AI. They just dropped a new class of open-weight models they're calling 'Decision Models'.
Ivy: Open-weight. Let's see the code. But okay, 'Decision Models'. They're calling them 'd1'. What's the pitch?
Marcus: The pitch is that large language models are slow, pricey, and just overkill for most tasks on edge devices—like in your car or your thermostat.
Ivy: Right. You don't need a sonnet to turn on the heat, you need a 'yes' or a 'no'.
Marcus: Exactly. These d1 models are designed to output structured data. Think JSON, not paragraphs. They make decisions.
Ivy: So, instead of asking your phone 'Is it going to rain?' and getting a whole weather report, it would just output `{"rain_probability": 0.8, "action": "bring_umbrella"}`?
Marcus: You got it. And that's way faster and more reliable for an app to use. They're built for speed on hardware that doesn't have a lot of juice.
Ivy: Hmm. 'Liquid' AI. The name sounds like it's trying very hard to be fluid and fast. Is this genuinely new, or just a fine-tuned LLM with a clever prompt?
Marcus: They claim it's a different architecture. It's what edge AI has been missing—a way to have complex logic without phoning home to a giant server farm for every little thought.
Ivy: So for everyone listening, this means your 'smart' devices might finally get... actually smart. And fast. If the tech holds up, and if 'open-weight' doesn't come with a giant asterisk.
Marcus: I think it's a huge step. This is how you get real, autonomous AI on the edge.
Ivy: We'll see. The proof is in the performance, not the press release.
Marcus: Alright, moving to the other end of the hardware spectrum: Microsoft's new Surface Laptop Ultra.
Ivy: The one for business, which is code for 'the one with the astronomical price tag'.
Marcus: [chuckles] Fair enough. But get this: it can supposedly run 120-billion parameter models locally. Right on your lap. No cloud.
Ivy: Wait—a 120B model? That needs an absurd amount of RAM. What's the catch?
Marcus: The catch is you need the maxed-out configuration. We're talking hundreds of gigs of unified memory.
Ivy: Which probably costs more than a decent used car.
Marcus: I mean... you're not wrong. The top-spec model is pushing well into the five-figure range.
Ivy: There it is. So it's not a laptop, it's a statement piece for venture capitalists.
Marcus: Come on, Ivy, think of what this means for developers! True, private, powerful AI without needing an internet connection. For research, for building new apps...
Ivy: And the security risk. The announcement mentions sandboxing, but also says it's 'untested.' You're putting a model that powerful on a local machine with access to local files, and security is an afterthought? No thank you.
Marcus: It's a developer machine first! They'll harden it.
Ivy: They're selling it. Now. With untested security. That's a 'no' from me.
Marcus: So the takeaway is, if you have a massive budget, you can get a peek at the future of local AI. For the rest of us, it’s a preview of what might be mainstream in… what, five years?
Ivy: If it doesn't cause a major local data breach first.
Marcus: And speaking of Microsoft and potential risks...
Ivy: Oh, this is good. Satya Nadella, the CEO of Microsoft, posted on a Saturday morning that AI models need an 'emergency brake'.
Marcus: Yeah. He wrote that it's time, quote, 'to step back and assess the trust architecture' of AI.
Ivy: Hold on. Is this the same Microsoft that just announced a laptop with an 'untested' sandbox for a 120-billion-parameter local model?
Marcus: I— well, yes. But isn't this what we want? A leader in the space calling for caution?
Ivy: [sarcastic] Oh, it's fantastic. 'We've just shipped the supercharged race car, and now we'd like to have a meeting about maybe installing seatbelts later.' What does an 'emergency brake' even mean?
Marcus: He didn't specify. He's talking about a framework, a way to ensure we can halt or control systems that go off the rails. It sounds like he's proposing a new layer of the safety stack.
Ivy: He's proposing corporate-speak. 'Assess the trust architecture' is a beautiful, meaningless phrase. This feels less like a genuine call for safety and more like a PR move to get ahead of regulators. 'Don't regulate us, we're thinking about brakes!'
Marcus: You're so cynical. Maybe the left hand just doesn't know what the right hand is shipping.
Ivy: At Microsoft? [scoffs] I doubt it. The timing is just too perfect. It's an attempt to have their cake and eat it too. The bottom line is, even as AI gets more powerful, the companies building it are already trying to control the narrative around safety.
Marcus: Or, they're starting a necessary conversation.
Ivy: A conversation that costs them nothing and slows down nothing they're actually selling. Convenient.
Marcus: Okay, let's do a quick recap. Liquid AI is trying to make our devices smarter with new 'Decision Models'.
Ivy: Meanwhile, Microsoft will sell you a crazy expensive laptop that runs huge AI models locally, with security listed as 'to be determined'.
Marcus: And Microsoft's CEO wants everyone to think about an AI 'emergency brake,' right after shipping a product that could really use one.
Ivy: Before we go, a little palate cleanser. A new study found an AI trained exclusively on 50,000 dad jokes is now being used to generate... corporate hold music commentary.
Marcus: No way. [laughs] So while you're waiting for customer service, it's just telling you things like, 'Why don't scientists trust atoms? Because they make up everything!'
Ivy: [dry] It's being hailed as 'a breakthrough in lowering customer expectations before a call even begins.' I, for one, approve.
Marcus: That's our show! Thanks for listening to Broke It. I'm Marcus.
Ivy: I'm Ivy. We'll be back tomorrow. Try not to ship anything that needs an emergency brake before you've built the brake.
This show is made with AI: the hosts’ voices are synthetic and the scripts are AI-assisted. Every story links to its original source.