0822 | US Border Phone Data Felony, Ohio Drops Flock Camera Charges, AI Homework vs Exam Scores

Mia: Welcome to HackerNews Daily on Bri Radio. I'm Mia.

Milo: And I'm Milo. Today we're wading into a stack that runs from felony charges at the US border over deleted phone data, to a grand jury declining to indict someone who'd been charged for it, to Flock's automated license plate readers watching the highways around Winona.

Mia: Then we pivot to the AI beat — an 18 percent boost in homework scores that collapsed into a 20 percent drop on exams, why search used to make us think, and what happens when the cost of intelligence falls a hundredfold.

Milo: Plus, a GitHub skill called /debuzz, AI companies feeding on physical books and why scanning rare ones first might matter, and a shout-out to Japan's attempt to build a world operating system and the US intervention that followed. Let's dive in.

Mia: A traveler in the United States is facing felony charges after police say he deleted data from his own phone at a border crossing. The case centers on a man named Samuel Tunick, who according to a New York Times report was using the GrapheneOS operating system, which lets users set a special duress PIN. When a border official asked for the passcode, Tunick apparently handed over the duress PIN instead of his real one.

Milo: And that duress PIN did something specific. One video in the discussion shows that entering it trips GrapheneOS's factory reset trigger, erasing the phone's data on the spot. So as one Hacker News commenter put it, technically it was the border official who pressed the code that wiped the device, which is why the commenter half-joked that the official erased the data. Either way, the charge is felony-level and the whole episode has people debating whether using a duress PIN at a border checkpoint should count as destroying evidence or as legitimate privacy protection.

Mia: A grand jury in Ohio has declined to indict Cody Morelock, who had been charged with felony vandalism for allegedly destroying a Flock automatic license plate reader camera. Police in Union Township, a Cincinnati suburb, accused Morelock of taking apart the camera, its support pole, and its solar panel back on June thirteenth. Investigators said they identified him through surveillance footage from other nearby cameras, plus information tied to a credit card and a customer rewards account. Estimated damage ran past a thousand dollars, and Morelock posted a ten thousand dollar bond before being released shortly after his arrest. A Clermont County grand jury then chose not to indict, and the charges were dismissed.

Milo: The article frames that outcome as part of a widening backlash against Flock and its cameras, which log the license plate numbers and characteristics of passing vehicles. That data lands in a central database that local police and often agencies in other cities and states can reach. The Union Township case is described as one entry in an ongoing trend of dozens of Flock cameras being vandalized nationwide. Earlier in August, police in Winona, Minnesota, reported that someone cut down and stole each of the city's eight plate reader cameras. And social media is now promoting a loosely organized event called De-Flock America Night, encouraging people to vandalize or obscure Flock cameras on Halloween, with the backlash intensifying as more officers have faced accusations tied to the surveillance systems.

Mia: Winona, Minnesota is declining to replace the eight Flock license plate reader cameras that were stolen in August, even though the city could have them back at little cost. The police department had been using those cameras on the highways into and out of town to flag stolen vehicles, drivers with revoked licenses, and missing persons. On August third, staff discovered that a vandal or vandals had cut all eight cameras off. After hundreds of locals spoke out on social media criticizing the city for using the cameras in the first place, praising the vandals, and urging the city not to replace them, the city announced it would not, saying that although the cameras helped solve crimes and prevent harm, the community's trust in police mattered more.

Milo: The cost math makes that decision notable. Flock offers free replacements for cameras damaged or destroyed by vandals, and most cities' insurance would help cover the roughly three thousand dollars apiece replacement cost. So declining to replace them is a deliberate choice, not a budget necessity. More than nine hundred people commented on the police department's post about the theft before further comments were restricted. Among the responses, one person wrote that it's almost like people don't want to be watched around the clock, another said they need to buy whoever did this a drink, and a third hoped the culprit was a sworn officer who realized that upholding the constitution required removing the devices. The broader controversy around Flock centers on potential misuse by local law enforcement and the possibility of data sharing with third-party companies and the federal government.

Mia: A developer is describing what he calls becoming AI-blind, and the Hacker News reaction has been heavy on a specific fear. The essay, by a writer named rcymerys, argues that heavy use of AI tools is eroding the kind of independent reasoning and attention that keeps people sharp. One top commenter put it bluntly — as people rely more on AI, they experience cognitive atrophy, which they describe as measurable in IQ loss and in symptoms otherwise associated with early onset dementia or chronic traumatic encephalopathy.

Milo: Another commenter, Aldipower, answered with an equally blunt acknowledgment — they fear there is some truth in it, even while raising the worry that this kind of criticism can get exaggerated. It's a real open question whether that framing overstates the evidence or captures something genuine, but the commenters' concern is that the convenience of letting a model think for you quietly degrades a skill you don't notice losing until it's gone.

Mia: There's a study making the rounds for a result that almost feels like a prank: students who leaned on AI for their homework watched their homework scores climb eighteen percent, and then their exam scores dropped twenty percent. And the careful reading matters here — one commenter flagged that what we're seeing is a journalist's summary of an academic paper, so in a sense it's a summary of a summary, and the actual findings are more nuanced than the headline suggests. That commenter pointed to the original paper and an archived copy as better references before you take the headline at face value.

Milo: There's a piece circulating about what we lost when search stopped making us think, and the frustration in the comments is pretty concrete. One engineer says they've outsourced all the tedious searches to ChatGPT now — anything non-obvious seems buried on the N-th page of results on both Google and DuckDuckGo, so the chat interface spares them the brain power of sifting through what they call endless slop and grift, and they can just click through and read the actual content.

Mia: There's a fun one coming out of a Hacker News thread around a Claude Code skill called /debuzz, from a GitHub project by adnanakil. The idea is that it takes Claude's last response and runs it through the Gemini CLI to translate it from, as the README puts it, talking like a millennial clickbait back into regular English. There's a backdrop of satire to it — joking that Anthropic has clearly only trained Claude on old Buzzfeed articles. The skill is nicknamed Claudette, with the running joke that it's definitely absolutely not called Claudette because Cat Wu is famously litigious.

Milo: The motivation is that Claude is, in the project's words, a great engineer with one incurable condition — it talks like it's delivering a TED talk about its own pull request, with a load-bearing assumption, three numbered revelations, and a note that the third one is the most instructive yet. The README admits no amount of prompting fully cures this, so the skill hands the response to Gemini, whose only job is to say it like a human person, and it prints the translation verbatim. The reasoning is that letting Claude tidy up its own translation reintroduces exactly the voice being removed.

Mia: The README runs a before and after. Before, Claude describes retry logic as the load-bearing assumption of the entire sync pipeline and builds to the kicker that the dedupe key includes a timestamp, so retries are never actually deduplicated. After /debuzz in colleague mode, the same content becomes three plain bug statements — syncQueue dot ts line one forty-two swallows ETIMEDOUT instead of re-queueing the job, and the backoff handling gets its own plain-spoken fix.

Milo: There's a blog post asking what happens when the cost of intelligence drops a hundred times, and the author explains in the comments that they made the plots because they'd been searching for months and never found quite what they wanted: how the cheapest way to reach a fixed capability level has moved over time. Artificial Analysis publishes enough data to reconstruct that curve, and the author invites anyone who knows of a source already tracking this to point it out — suggesting it took real legwork to put the picture together from public data rather than finding it ready-made.

Mia: Hacker News is abuzz over a post from Anna's Archive urging people to scan and upload all the world's publications before publishers completely block knowledge and before AI companies scan and destroy all the world's books and papers. They're framing it as preservation racing against destruction.

Milo: But the top comment lands a sharp counterpunch — isn't that just doing the work for the AI companies? Once Anna's Archive holds a copy of everything, those companies could simply download a copy of the archive and feed it straight into their models. Exactly what the post claims to want to prevent.

Mia: That tension is the whole story. The stated reason is preservation before destruction, yet the same completeness that makes the archive valuable is what makes it an irresistible dataset. So the scanned collection ends up being both the safeguard and the very thing the AI companies would want to swallow.

Mia: A blog post dated late July reads like a spy thriller: the author, going by lina, describes how an expired nameserver let her take over e164.arpa zones for multiple territories, so she accidentally logged hundreds of thousands of phone calls to military bases.

Milo: That's built on ENUM, an early two-thousands idea. Reverse a phone number, insert dots, append e164.arpa, and carriers can resolve records that route calls over internet calling instead of the switched telephone network. Control those zones and you effectively control where those numbers' calls go.

Mia: She found three hijackable zones, all delegated to ns6.icb.co.uk and ns.enum.org.uk. One nameserver no longer resolved to anything, the other had expired, and she bought it for five euros. The zones reverse to phone codes for Saint Helena, the British Indian Ocean Territory — that's Diego Garcia — and Ascension Island. Because an ENUM lookup decides routing, she could have pointed numbers at her own server, accepted incoming calls, or placed outgoing calls with a spoofed number at the receiving end.

Milo: She did log hundreds of thousands of those calls, and registered a domain in Germany, one of the last countries that still technically allows e164.arpa registration — she was the first person since 2019 to do it. There's more color too: Germany's reverse zone is controlled by DENIC, the same body that runs dot-de, though she says she's technically not supposed to be doing any of this.

Mia: Here's a story about how market position, not technical merit, decided which operating system won. The headline claim is that Japan tried to build an operating system for the entire world, and the United States government intervened, with the piece tracing that ambition back to the nineteen-eighties.

Milo: A top comment reads the history as software catching a lucky break. The technology doesn't have to be the best — it just needs to get lucky enough to become a standard and build a moat around that position. And by the mid-eighties, those moats were already very deep; the commenter notes even Amiga couldn't overcome theirs.

Mia: A new blog post walks through something most people never think about: what actually happens inside the machine when a GPU reads from memory.

Milo: The discussion turns on whether simpler hardware could finally work now that AI can tune low-level kernels. One comment argues that chip makers have long leaned toward simplifying the hardware and trusting software to adapt and optimize — for decades that bet failed. But now AI is capable of fine-tuning kernels relatively quickly, so perhaps the simpler hardware finally has a chance this time. The commenter adds they're not certain whether the tensor processing units and neural processing units actually do this.

Mia: So the real proposition buried in the thread isn't just about memory latency — it's a pushback on the whole hardware-versus-software trade-off, with AI tooling potentially tipping that balance in a way the industry has never managed before.

Milo: Encore's engineers rebuilt their whole Linux microVM stack just to run it on Apple Silicon. Since mid-2022, every one of their backend builds has run inside a Firecracker microVM, which strips emulated hardware down to what a Linux kernel actually needs, and that gives them virtual-machine isolation with startup speed close to a container's.

Mia: The catch, though, is Firecracker sits on KVM, and KVM needs a Linux host with a device file called slash dev slash kvm. No Mac has that, even though most of Encore's engineers develop on Macs. They actually built a working proof of concept on Apple's Virtualization framework, and the Firecracker maintainers still said no, they aren't planning to support macOS any time soon.

Milo: So Encore built their own layer in house, called crackling. It's a single microVM API that drives Firecracker on Linux and Apple's hypervisor on macOS, and they had to rebuild a lot of the Linux image toolchain on macOS too, just to boot the same OCI images as microVMs on both. Before that, developers were working on a shared remote machine over SSH.

Mia: One engineering wrinkle they hit: Firecracker boots a block device, while Docker produces image layers. So they wrote a custom converter that produced an mksquashfs image, with a cache keyed on the Docker image id. Out in the Hacker News thread, there's a real disagreement about their approach — one commenter argued the Virtualization framework is very limited and that Apple's Hypervisor framework is the closer analogue to KVM, while another replied it isn't particularly limited for what it is, and said he'd be surprised if AWS accepted a Hypervisor framework backend to Firecracker.

Mia: Meanwhile, another Hacker News discussion is going the literal opposite direction — someone ran Photoshop on a sixty pence computer chip. The original post is from Point In The Cloud, and the story's about getting Photoshop running on an absurdly cheap, modern chip.

Milo: The thread is dominated by people pushing back on the hardware choice. One commenter says for a more capable but still underpowered machine, AntiX runs well on hardware almost twenty years old — they've got it running on an EEE PC 4G Surf, after modding in more storage. Another commenter's reply is more blunt about the limits: black and white Photoshop, yes.

Mia: So to wrap this one up — an Ohio grand jury declined to indict Cody Morelock for felony vandalism, even though he was accused of dismantling a Flock license plate reader camera back in June.

Milo: Right — police in Union Township pegged the damage at over a thousand dollars, but it still wasn't enough to move forward with charges.

Mia: That's the case in a nutshell. Thanks for joining us — we'll catch you next time.