Ivy: So, the guy who basically reinvented the modern hotel now works for OpenAI.
Marcus: I'm Marcus.
Ivy: And I'm Ivy.
Marcus: And you're listening to Broke It. It’s Wednesday, November 11th, and we’re breaking down the three biggest stories in AI.
Marcus: OpenAI just made a very unexpected move into the travel industry.
Ivy: New York's big AI safety law is officially here, and the top model makers are about to face a whole new level of oversight.
Marcus: And we go hands-on with Meta's new Muse Ember model... let's just say first impressions are... mixed.
Marcus: Okay, so OpenAI just acquired Odessia, a travel startup from Sonder founder Francis Davidson.
Ivy: An AI company buying a travel company... I did not have that on my 2026 bingo card.
Marcus: But it makes perfect sense! Imagine a GPT that doesn't just find flights, but plans your entire multi-stop trip—visas, hotels, dinner reservations... the works. Seamlessly.
Ivy: Dream on. OpenAI has said nothing about its plans. This could just be an acqui-hire for Davidson's expertise.
Marcus: You don't buy a whole company just for one guy, even a big name. They're buying the team, the vision...
Ivy: TechCrunch calls it an 'early-stage startup.' What team? What vision? It sounds like they bought a name and a slide deck. We don't even know what they paid.
Marcus: They paid enough to get one of the sharpest minds in hospitality. Sonder completely changed the rental game. Applying that logic to AI travel is a trillion-dollar idea.
Ivy: A trillion dollars... for an AI that can book a hotel without making up the check-in time? Right.
Marcus: Yes! That is the point! Getting the details right is the hard part. This shows they're serious about building real-world agents.
Ivy: So what does this actually mean for someone planning a vacation next year?
Marcus: It means your next trip might be planned by ChatGPT—and it'll actually work. No more wasting hours on Kayak and TripAdvisor.
Ivy: Or it means nothing changes for five years while they burn cash in R&D. We just don't know. It’s a press release, not a product.
Ivy: Well, it looks like the regulators are finally catching up. New York's big AI safety law is now in effect.
Marcus: Ugh. Here we go. More red tape.
Ivy: Or... more accountability. It forces 'major developers'—OpenAI, Anthropic, Google, all the big names—to register with the state.
Marcus: Register? What is this, the DMV for LLMs? What does that even accomplish?
Ivy: It makes a list of who's building this stuff. Plus, they have to disclose safety practices and—here's the kicker—report any 'critical incidents.'
Marcus: And what counts as a 'critical incident'? A model giving a bad answer? A server going down?
Ivy: The law is specific. We're talking major security breaches, real evidence of systemic bias, or the model being used for large-scale fraud. You know, things we actually want to know about.
Marcus: Who decides what's 'critical'? A panel of politicians in Albany? They don't know the first thing about how these models work. This is just going to slow everything down.
Ivy: When you're building something this powerful, maybe slowing down isn't the worst idea. What's the alternative? Just... trust them? After everything we've seen?
Marcus: I hear the fines for not complying are pretty steep.
Ivy: Good. That’s called an incentive. So, what does this mean for developers building on top of these models?
Marcus: It means your model provider has a new layer of compliance costs and legal risk, which they will absolutely pass on to you.
Ivy: It also means you might get a heads-up before a model you depend on has a meltdown. Seems like a fair trade-off.
Marcus: Alright, let's talk new models. Meta just dropped Muse Ember 1.0.
Ivy: It's on LLM Gateway, and the first impressions are... not great.
Marcus: Look, it's a 1.0 release. Of course it's going to be inconsistent. You have to give these things time.
Ivy: There's 'needs time to mature,' and then there's just plain 'not ready.' The biggest complaint isn't even performance, it's the missing documentation.
Marcus: What kind of docs are we talking about?
Ivy: The kind that tell you how to fine-tune it, its failure modes, context limits, biases... you know, the basic stuff you need before you bet your company on it.
Marcus: That's fair. But it's Meta. They'll release it, they always do. It's probably just a few weeks behind.
Ivy: Probably? You can't build a product on 'probably.' It's like they handed you a Ferrari engine with no manual. Sure, it looks cool, but good luck putting it in a car.
Marcus: So, what's the takeaway for developers?
Ivy: Play with it on LLM Gateway. See what it can do. But do not put it in a production app until that documentation appears. For now, it's a toy, not a tool.
Marcus: I still think it's exciting. A new open-source model from Meta is always a win for the ecosystem.
Ivy: An exciting toy, maybe. Let's just hope the manual comes out soon.
Marcus: Alright, let's recap the big three. OpenAI is getting into the travel business... maybe.
Ivy: New York is rolling out the first real AI safety rules for Big Tech... definitely.
Marcus: And Meta's new Muse Ember is out, but maybe wait for the instruction manual before you build anything with it.
Ivy: And before we go, a new study tested an AI on generating Thanksgiving recipes.
Marcus: Oh yeah? And? Did it solve the dry turkey problem?
Ivy: Not exactly. One of its top suggestions was for 'gelatinous mushroom gravy.' Another was a pumpkin pie made with mashed potatoes.
Marcus: Okay, my job as family chef is safe for another year. Gelatinous mushroom gravy...
Marcus: That's our show! We'll be back tomorrow with more on Broke It.
Ivy: See you then. And Marcus, please, don't make the gravy.
This show is made with AI: the hosts’ voices are synthetic and the scripts are AI-assisted. Every story links to its original source.