Sam: So Meta hired the same experts Big Tobacco used to deny addiction. And we're supposed to be surprised?
Kai: [beat] Whoa. Okay. I'm Kai.
Sam: And I'm Sam.
Kai: And this is Open Source, Unstarred. It's Wednesday, August 19th, and we have a lot to get to.
Kai: Starting with Meta's massive lawsuit, which is drawing some... unflattering comparisons to an industry we hoped was history.
Sam: Followed by a stunning report on what happens to kids' health when you actually get cars out of the city.
Kai: And we'll get into a new AI chip so big, they literally could not fit another transistor on it. We're talking wafer-scale.
Kai: Okay, so The Economist is reporting on Meta's blockbuster trial, and the parallels they're drawing to Big Tobacco... they're staggering.
Sam: Staggering, but is anyone really surprised? Like I said off the top, this confirms they're running the exact same playbook, right down to hiring the same consultants to sow doubt.
Kai: That's the part I can't get over. It's so brazen. It’s one thing to have your strategy compared to tobacco companies, it’s another to literally share expert witnesses.
Sam: Why reinvent the wheel when the old playbook worked for decades? The core of the lawsuit is that Meta knew its apps were harmful and addictive for teens, hid the research, and then tried to discredit anyone who pointed it out.
Kai: The comments are going wild. A lot of developers are having a real crisis of conscience, saying things like 'I helped build these engagement loops.' There's a real sense of culpability.
Sam: Good. They should. If you're a developer working there, this is when you have to ask: am I building a useful tool, or a digital slot machine? Are my KPIs about user well-being, or just screen time?
Kai: But isn't the whole industry built on engagement metrics? How do you even escape that?
Sam: You don't escape it, you challenge it. This lawsuit, if it works, could force their hand. We could see regulations mandating different metrics, limiting certain features, demanding real transparency. It could fundamentally change how this software gets built.
Kai: So the fallout could actually be... good? A forcing function for building healthier products?
Sam: [dry] After billions in fines and a decade of damage, sure. Let's call it a win.
Kai: [sighs] Okay, on to something more hopeful. A BBC report finds that kids' lung health in London's ultra low emission zone has seen a 'stunning' recovery.
Sam: The data is pretty remarkable. Kids inside the zone have about 5% larger lung capacity, which basically reverses the damage from air pollution.
Kai: Five percent! That's huge! And this feels like a clear win for tech, right? The sensors, the cameras, the data analysis... you implement the policy, you see the result. It's a perfect feedback loop.
Sam: Unambiguous? Kai, this is the same tech that privacy advocates have been warning about for years. A massive, city-wide surveillance network that tracks every vehicle's movement. Are the health benefits worth living in a panopticon?
Kai: But it's for a public good! It's not tracking people, it's tracking polluting vehicles. And the result is literally saving children's lives. Isn't that a worthwhile trade-off?
Sam: It's a question we should be asking, not just accepting. For developers, this is the catch-22 of 'tech for good.' You think you're working on a health project, but you're also building a surveillance tool. You have to see both sides.
Kai: I see the point, but I'm still putting this one in the 'win' column. The data is just too good.
Sam: The data is good. The implementation details are... complicated.
Kai: Alright, let's talk about hardware. Big hardware. Cerebras just announced their new CS-4 chip, and it is a beast.
Sam: The CS-4 'Wafer-Scale Engine'. They're not shy with the marketing, are they?
Kai: [excited] Why should they be? Sam, this thing has four trillion transistors. Not a typo. An entire silicon wafer turned into one massive chip.
Sam: Four trillion transistors, and probably draws enough power to dim the lights in a small city. What do their benchmarks say?
Kai: They're claiming a 10x performance leap over the previous generation for training large models. And get this: they're selling it in clusters of up to 2,048 systems. That's a quarter-exaflop of AI compute. Imagine what you could train with that!
Sam: '10x performance' on what, exactly? Their own proprietary benchmark that no one can replicate? This is classic hardware marketing. Big, headline-grabbing numbers with very little context.
Kai: You are impossible to impress. This is bleeding-edge engineering! It enables research that is physically impossible on commodity hardware.
Sam: It's an amazing piece of engineering, I'll give you that. But for 99.9% of developers listening, it means nothing. You'll never touch this machine. The models trained on it, though? They'll just concentrate more AI power in the hands of the few who can afford a nine-figure supercomputer.
Kai: Or it pushes the whole field forward and those advances trickle down, like they always do. The first car wasn't for everyone, either.
Sam: Let's check back when a CS-4 costs less than a small nation's GDP. For now, it's a beautiful, exotic, and largely irrelevant piece of tech for the average dev.
Sam: Okay, let's recap. Meta's legal strategy is to cosplay as Big Tobacco.
Kai: Cleaner air is literally healing children's lungs, but Sam is worried about the cameras.
Sam: And Cerebras built an AI chip the size of a dinner plate that you'll never, ever get to use.
Kai: [laughs] Perfect. Okay, before we go, I found a fun one on GitHub. It's a Prettier plugin called `prettier-plugin-lunar`.
Sam: Don't tell me. It formats your code based on the current phase of the moon?
Kai: Exactly! Wider spacing during a full moon for 'clarity', more compressed during a new moon to be 'concise'. It's completely useless and I absolutely love it.
Kai: And that's our show. As always, links to everything we talked about are in the show notes.
Sam: We'll be back tomorrow. In the meantime, maybe check who your expert witnesses have worked for in the past.
This show is made with AI: the hosts’ voices are synthetic and the scripts are AI-assisted. Every story links to its original source.