0:00 Xiaomi just open-sourced an AI model that beats every other open-source model out there.0:05 It’s called MiMo-V2.6-Pro, and it’s a genuine leap forward—which is exactly the kind of thing we were talking about last week with all the moves in open-source AI and hardware.0:17 It feels like the floor for what’s possible is rising every single day.0:22 But this week, the most important stories weren’t just about making AI smarter.0:28 They were about what happens when AI gets woven so deep into our lives that we can’t say no.0:34 And that’s a much more complicated conversation.0:37 Let’s start with the headlines.0:39 First, something completely different, and honestly, a real gut punch for anyone who loves space exploration.0:47 NASA’s Mars Sample Return mission is officially dead.0:50 Congress backed the Trump administration’s budget cuts in the 2026 spending bill, and just like that, the plan to bring pieces of Mars back to Earth is over.1:01 For years, this has been the top priority for planetary scientists—the chance to analyze Martian rocks in labs here, with equipment you could never dream of sending on a rover.1:13 And now… it’s just gone.1:15 A decades-long dream, shelved.1:17 Back on Earth, we’re seeing the second-order effects of AI assistants becoming common for developers.1:24 The team at Linear, the project management tool, wrote about how they had to completely re-engineer their testing systems.1:32 Why?1:32 Because with AI helping write code, their developers were shipping SO much more, so much faster.1:39 Their number of test suites nearly quadrupled this year alone.1:43 And if they hadn’t done anything, the wait time for a pull request would have ballooned.1:49 Instead, they moved workloads, switched compilers, and optimized everything they could.1:55 The result?1:55 Wait times actually went DOWN, from over six minutes to just over five.2:00 It’s a perfect little case study of how adopting one technology forces you to level up everything around it.2:08 And speaking of understanding the tech, a visual explainer of the Transformer architecture got a ton of attention.2:15 It’s the core design behind basically all the big models—ChatGPT, Gemini, you name it.2:21 The fact that hundreds of people were geeking out over diagrams of token embeddings and multi-head self-attention just shows you how much of a hunger there is to understand what’s actually going on under the hood.2:36 Now, for the stories that really define the week.2:39 There are three of them, and they’re all connected by this thread of control.2:44 Who has it, and who’s losing it.2:46 First, a massively popular essay by Colin Breck called “I don’t want to read what you didn’t write.” It’s a critique of how AI is being used in the workplace, and it is scathing.2:59 He says people who never used to write are now generating these enormous, detailed design documents with AI, and, in his words, they are “all unreadable.” Then, a new, and frankly, kind of terrifying term popped up: “spymarks.” An article on Brand.io laid out how companies can embed invisible, hidden tracking signals in images, audio, even text.3:22 These aren't just watermarks to prove ownership; they can secretly encode things like your IP address or a unique user ID, making anything you create or share traceable back to you.3:35 And finally, the one that ties it all together.3:38 Apple—the company that built its brand on privacy—is reportedly removing the user’s ability to completely opt out of its “Apple Intelligence” data reporting in the new macOS.3:50 Even if you said no before, the new system just… keeps collecting.3:54 So you have AI making our writing worse, our content secretly traceable, and our operating systems ignoring our explicit desire for privacy.4:04 See the pattern?4:05 Okay, let’s go deep on this, because it’s important.4:08 Let’s start with the writing.4:10 Colin Breck’s essay really struck a nerve on Hacker News.4:14 He describes this new wave of AI-generated workplace documents as "inhumane." That’s a strong word, but when you read his argument, you start to get it.4:24 He says a design document used to be a tool for building consensus.4:29 You wrote it to persuade your colleagues, to lay out your thinking, to be vulnerable, and to invite feedback.4:36 It was a social object as much as a technical one.4:40 Now, he says, that’s being replaced by something else entirely.4:44 He writes, “it is a summary made by a machine, in exhausting detail, without context or perspective.” It’s unreadable because it wasn’t written for a reader.4:55 It was assembled.4:56 It has all the facts but none of the meaning.4:59 It doesn’t tell you why a decision was made, what trade-offs were considered, or what the author truly believes.5:07 It’s just a pile of structured text.5:09 And the most incredible, almost comical example he gives?5:13 He says, “I had someone use AI to summarize my comments on their proposal as their response to me!” I mean, wow.5:20 It’s like, thanks for the feedback, I’ve had my robot repackage your thoughts and present them back to you as if they’re my own.5:29 It’s not just lazy; as he points out, it’s offensive.5:33 It short-circuits the whole process of thinking and communication.5:37 This isn’t just about being a bad writer.5:40 This is about the erosion of a fundamental workplace skill: making a case.5:45 And where have we seen this before?5:47 Well, in a way, it's the logical endpoint of decades of "boilerplate." Think about how legal documents or corporate memos started to become these copy-pasted monstrosities that nobody really read, but everyone was forced to use.6:03 AI is like boilerplate on super-steroids.6:06 It takes the human element, the perspective, the why, and sands it all down into a smooth, featureless, and ultimately meaningless surface.6:15 The discussion on Hacker News had some pushback, of course.6:19 People argued that AI is a great tool for a first draft, that it helps overcome writer’s block.6:26 And that’s true!6:27 But Breck’s point is about where it ends.6:30 If the AI-generated draft becomes the final product, and if that becomes the norm, we lose something critical.6:37 We stop building consensus and we start… managing output.6:41 And a culture that only manages output, without understanding the thinking behind it, is a culture that’s incredibly brittle.6:49 So that's the loss of control over what we create.6:53 Now let's talk about the loss of control over what's done to us.6:57 This brings us to “spymarks.” This is a new word for an old idea, but the technology has taken a quantum leap.7:05 A traditional watermark is usually visible.7:08 You can see the faint logo on a stock photo.7:11 Its purpose is overt.7:12 A spymark, as the Brand.io article defines it, is a hidden signal that makes your work traceable without your knowledge or consent.7:21 It’s steganography, but industrialized and deployed at scale.7:25 The article uses Google’s SynthID system as a prime example.7:29 It can embed a 136-bit payload into a 512 by 512 pixel image.7:33 Now, 136 bits might not sound like much.7:36 But the article breaks it down: that’s enough for a 64-bit database identifier, with 72 bits left over for error correction.7:44 Let's pause on that.7:46 A 64-bit number.7:47 Do you know how big that is?7:49 It’s eighteen quintillion.7:51 That is more than enough to assign a unique ID to every single person on Earth, multiple times over.7:57 So you share an AI-generated image, or maybe just a regular photo that’s been processed by a cloud service.8:05 And hidden in the pixels, imperceptible to your eye, is a unique number that could be tied back to your account, your IP address, your name.8:14 The argument for this technology, which you’ll hear from its proponents, is about provenance and fighting misinformation.8:22 Imagine being able to tell if an image was generated by a specific AI model, or if a news photo is authentic.8:30 That sounds good, right?8:31 But the analogy here breaks down.8:34 This isn't like a signature on a painting.8:37 It's more like every book in the library having a secret tracker that reports back to the publisher every time you open it, and which page you're on.8:47 The privacy implications are enormous.8:49 Who stores these IDs?8:51 Who has access to the database that links the ID to a person?8:55 Can it be used to track political dissidents?8:58 To monitor what content people are creating and sharing?9:02 The answer to all of those is almost certainly yes.9:05 We’ve seen this pattern before with web tracking.9:09 First, there were simple cookies, which felt kind of invasive.9:13 Then came browser fingerprinting, which was way more persistent and harder to block.9:18 Each step was a gradual escalation, an erosion of anonymity and control.9:23 Spymarks feel like the next logical step in that progression.9:28 It moves tracking from the browser and the network right into the content itself.9:33 The files become their own surveillance devices.9:36 And because it’s invisible, there’s no way for an average user to know it’s even happening.9:42 There’s no consent.9:44 And that lack of consent brings us to Apple.9:47 For years, Apple’s entire brand has been built on the foundation of privacy.9:52 “What happens on your iPhone, stays on your iPhone.” They relentlessly attacked Google and Facebook for their data-hungry business models.10:01 They positioned themselves as the user’s champion.10:05 Which is what makes the news about macOS 27 so jarring.10:08 According to a post on dbushell.com that got a lot of traction, Apple has removed the explicit “no” toggle for its Apple Intelligence data reporting.10:19 Users who had previously opted out are finding that background processes are still running, still collecting data.10:26 The ability to give a clear, unambiguous refusal has been taken away.10:31 The author of that post does not mince words.10:34 He says, “The AI industry is vile...10:37 You will use it.10:38 You cannot look away, it will be stuffed down your throat and if you say ‘no’ they’ll cut out your tongue.” That’s intense.10:46 But it captures the feeling of helplessness that a lot of people have right now.10:51 You bought the device.10:53 It’s your computer.10:54 But you don’t seem to have the final say over what it does.10:58 The pattern here is the classic boiling frog.11:01 A company offers a new feature.11:03 At first, it's opt-in.11:05 Then it becomes opt-out.11:07 Then the opt-out button gets hidden in a series of confusing menus.11:11 And finally, the button disappears entirely.11:14 We’ve seen this with social media privacy settings, with ad tracking, with location services.11:20 And now we’re seeing it with OS-level AI integration.11:24 The defense, which you can find in the Hacker News comments, is that these AI features are now so deeply integrated into the operating system that you can’t just turn them off without breaking things.11:38 That a personalized, predictive experience requires this data.11:42 But that’s a choice about system design.11:45 Apple chose to design it this way.11:47 They could have prioritized preserving that explicit “no.” They didn’t.11:52 So you put the three stories together.11:54 An AI that writes for us, but in a way that erodes meaning.11:58 An AI that embeds secret trackers in our creations.12:02 And an AI that runs on our computers whether we want it to or not.12:06 It paints a pretty clear picture.12:08 The push is towards a future where the friction of using AI is zero, but the cost is our own agency.12:15 So what does this all set up?12:17 This week, we got a perfect snapshot of the two parallel tracks of AI development.12:23 On one hand, you have incredible new capabilities, like Xiaomi's MiMo model, pushing the boundaries of what’s possible, especially in the open-source world.12:33 That’s the story of progress, of power, of sheer technical achievement.12:38 But on the other hand, you have this steady, quiet erosion of human control.12:43 It’s happening in our creative work, in our privacy, and in the very devices we own.12:49 And that’s the story that I think matters more in the long run.12:53 The fight over the next few years isn’t going to be about who has the highest benchmark score.13:00 It’s going to be about who gets to set the terms.13:03 It's about whether we build these systems with non-negotiable off-switches, with clear and explicit consent, and with a deep respect for human context and perspective.13:15 Because a tool you can’t turn off isn’t really a tool.13:18 It’s your new boss.
