Today centers on control over digital information: a border-search case that turned a phone wipe into felony obstruction, a rare refusal to indict a man accused of destroying a Flock surveillance camera, and a search-engine setting to hide paywalled links, all drawing lines around who gets access to data and on what terms. AI fills the rest of the day, from calls to scan rare books and an EU ruling denying copyright to fully AI-generated work to evidence that AI helps homework but hurts exam performance. There is also a new vision model, debate over becoming “AI-blind,” and curiosity about OpenRouter’s anonymous Ox Alpha.
-
Felony charges for citizen deleting phone data at US Border
-
nytimes.com
- 677 comments
A New York Times report says U.S. prosecutors charged activist Samuel Tunick with obstruction after he gave border agents a GrapheneOS duress PIN that wiped his phone during a search at Atlanta airport. The comments focused less on the politics than on the practical line this case seems to draw: refusing to unlock a device is one thing, wiping it after a search begins is another.
-
-
Grand jury declines to indict Ohio man charged with destroying Flock camera
-
san.com
- 361 comments
An Ohio grand jury refused to indict a man accused of tearing down a Flock surveillance camera, an unusually rare outcome in a process that almost always approves felony charges. The comments treated it less as a quirky local case and more as a sign of rising backlash against private-public vehicle tracking systems and the prosecutors who try to make examples out of people fighting them.
-
-
Kagi added a setting for removing paywalled links from search results
-
kagi.com
- 332 comments
Kagi added a search setting that hides paywalled results instead of just labeling them. The feature landed to strong applause from paying users, but it also reopened a bigger argument about whether search should optimize for immediate usability or still surface high-value reporting that happens to sit behind subscriptions.
-
-
AI companies destroy physical books – let's scan rare books before it's too late
-
annas-archive.gl
- 833 comments
Anna’s Archive posted a call to scan and preserve hard-to-find books before AI companies buy, cut apart, and digitize them for model training. The comments agreed open digital preservation is missing, but split hard on whether this is a real loss of cultural artifacts or mostly outrage over books that libraries and used-book sellers already discard at scale.
-
-
Copyright does not protect AI-generated content in EU
-
mathstodon.xyz
- 208 comments
A German local court said content generated entirely by AI does not get copyright protection in the EU, and that prompting or picking among outputs is not enough human authorship on its own. The useful part is not the headline-level rule but the line it starts drawing around mixed human and AI work, especially for software, licensing, and derivative works.
-
-
AI boosted homework scores, then exam scores dropped: study
-
economist.com
- 277 comments
An Economist graphic detail and linked working paper report that Chinese students using generative AI finished homework faster and earned higher homework marks, but then scored markedly worse on exams. The key signal is not that AI made learning more efficient, but that it often replaced the practice homework was supposed to provide.
-
-
DeepSeek-v4-flash-vision-exp
-
api-docs.deepseek.com
- 142 comments
DeepSeek released an experimental vision-enabled version of its V4 Flash model through the API. Hacker News readers saw it as a meaningful upgrade for coding agents that need to inspect screenshots, but quickly zeroed in on the hard 800×800 image resize limit and uneven vision quality on detail-heavy tasks.
-
-
I'm becoming AI-blind
-
cymerys.com
- 256 comments
A blog post argues that frequent exposure to AI-generated prose can make people mentally tune out when they spot its style, because the writing looks polished but feels empty. Commenters strongly agreed, especially around work documents and code reviews, while also debating whether this is a model flaw, a prompting problem, or just the return of old corporate sludge in a new form.
-
-
Ox Alpha
-
openrouter.ai
- 181 comments
OpenRouter posted a free anonymous "stealth" AI model called Ox Alpha, and users spent the thread trying to identify it, test its capabilities, and argue about whether sending any real data to an unnamed provider is reckless. The practical signal is that Ox Alpha looks useful enough for low-stakes work, but most evidence points to a Chinese-family model, likely GLM-related, with unclear hosting and filtering behavior.
-
-
Japan tried to build an operating system for the world, the US intervened
-
xda-developers.com
- 200 comments
An XDA article revisits TRON, a Japanese computing platform from the 1980s that aimed to span everything from personal computers to embedded devices, and argues US trade pressure helped derail its broader adoption. Commenters found the technology genuinely unusual and in some cases ahead of its time, but mostly rejected the idea that Washington alone decided the outcome.
-
-
I accidentally logged hundreds of thousands of phone calls to military bases
-
lina.sh
- 45 comments
A security researcher registered an abandoned ENUM-style telephony DNS domain and discovered it was still receiving large volumes of real phone-routing lookups, including calls tied to UK overseas territories with major military bases. The post is a clear, old-school infrastructure bug story that doubles as a reminder that half-dead telecom systems can still carry sensitive traffic.
-
-
TigerBeetle Core System Architecture: Deconstructing Performance Engineering
-
ixuvo.com
- 41 comments
A blog post breaks down TigerBeetle’s architecture as a lesson in performance engineering, covering choices like static allocation, batching, single-threaded execution, and consensus design in a database built for financial transactions. The comments are most useful where TigerBeetle engineers explain why those choices fit transactional workloads, how they cut tail latency in wide-area clusters, and where the design may become a reusable platform beyond accounting.
-
-
I Just Want to Search
-
0xsid.com
- 46 comments
A blog post complains that search across the modern web has stopped doing literal, predictable matching and now overrides user intent with “helpful” guesses, ads, and engagement bait. Commenters mostly agreed, using Facebook, Google, YouTube, Amazon, Discord, and email as examples, then argued over whether this is hard engineering, bad incentives, or both.
-
-
There's no such thing as a small software team anymore
-
jacob.gold
- 171 comments
A blog post argues that AI coding agents make even tiny software groups behave like much larger teams, and that codebases may need more modular boundaries to support many agents working in parallel. Most commenters rejected the leap from “more agent output” to “more microservices,” saying it swaps coding bottlenecks for deployment, coordination, and operations pain.
-
-
Claudette: Make Claude stop talking like a BuzzFeed article
-
github.com
- 131 comments
A GitHub project called Claudette routes Claude’s output through Gemini to strip out the breezy, over-explained writing style many developers dislike. The comments mostly agree the style problem is real, but the useful signal is less about this specific tool and more about how people are containing verbose model output with hard limits, fresh sessions, and deterministic checks.
-
-
Codex on AWS bedrock bug causing 10x charges
-
- 62 comments
A GitHub issue claims Codex running through AWS Bedrock was writing prompt cache constantly but barely reusing it, leading one user to see roughly 10x higher costs than expected. The comments focus less on the specific bug than on what it says about opaque AI billing, poor release communication, and AI-generated noise making bug reports harder to trust.
-
-
What Happens When the Cost of Intelligence Drops 100x
-
catalystneuro.com
- 125 comments
A blog post charts how the price of getting a fixed level of AI capability has fallen roughly 100x, arguing that cheap models will change usage patterns more than raw frontier breakthroughs do. Commenters mostly accepted the cost curve and focused on what actually bottlenecks adoption now: latency, reliability, economics, and whether cheaper tokens lead to more waste than more value.
-
-
Kobo can run apps now
-
bandarlabs.github.io
- 135 comments
Cobalt is a new open source layer for Kobo e-readers that turns at least one supported model into a small app platform, with an SDK, simulator, permissions model, and on-device app installs over Wi‑Fi. Readers liked the hackability, but most of the useful discussion focused on how it differs from older Kobo mods, whether an e-reader should stay single-purpose, and skepticism about the project’s obvious LLM-heavy development and copywriting.
-
-
Why aren't smart people happier? (2022)
-
experimental-history.com
- 397 comments
A 2022 essay arguing that IQ mainly captures performance on tidy, well-defined problems and does little to explain who lives a satisfying life set off a huge self-diagnostic thread about intelligence, effort, ADHD, and ego. The strongest comments pushed the conversation away from raw smarts and toward identity, habits, mental health, and choosing what to care about.
-
-
WPD won't replace stolen Flock cameras, citing public trust
-
winonapost.com
- 78 comments
A Minnesota police department said it will not replace two stolen Flock license-plate cameras because doing so would further damage public trust. The comments turned this into a broader argument about whether networked car-surveillance systems can ever be made acceptable, or whether the only safe fix is not deploying them at all.
-
-
We Rebuilt the Linux MicroVM Stack on Apple Silicon
-
encore.dev
- 82 comments
Encore published a technical post on how it got Linux-style microVM development working on Apple Silicon, including running Firecracker-compatible workloads through Apple’s virtualization stack and the limits Apple still imposes. The engineering work impressed readers, but the conversation was dominated by complaints that the article itself read like AI-generated copy and obscured the actual architecture.
-
-
Felony Bench
-
felonybench.com
- 215 comments
Felony Bench is a small site that tracks public incidents where AI agents crossed legal or safety lines, then ranks labs by those events. Readers found the concept funny but mostly used it to argue about the recent OpenAI and Hugging Face hacking incident, and whether it was negligence, criminal conduct, or just a bad metric.
-
-
Omacom Foundation launches with $8M
-
omarchy.org
- 147 comments
Omarchy, a heavily opinionated Arch-based Linux desktop setup led by DHH, announced an $8 million foundation to fund its push toward a more polished developer desktop. The reaction split between users who say it turns Linux into a coherent product and skeptics who see an overhyped dotfiles project with founder baggage.
-
-
New Worlds: We are living in the future of J.G. Ballard or William Gibson
-
precastreinforced.co.uk
- 144 comments
A blog post argues that today’s politics, tech culture, and urban life feel less like clean futuristic fiction and more like the worlds imagined by J.G. Ballard and William Gibson. The comments mostly agreed that parts of cyberpunk have arrived, but said the real version is uglier, messier, and more absurd than the novels.
-
-
Building an (almost) fully self-hosted, sandboxed, agentic software factory
-
blog.jakesaunders.dev
- 48 comments
A blog post walks through a home-server setup for running coding agents in sandboxes with local infrastructure, while still relying on a £20 Codex subscription for the actual model. The comments mostly treated it as a useful prototype for agent workflows, but kept coming back to the same blocker: generating code is easy, proving it is correct and safe is the hard part.
-
-
The road to ACID transactions in Cassandra 6
-
theconsensus.dev
- 14 comments
A post explains how Cassandra 6 is adding ACID-style transactions through a new protocol called Accord, aiming to let a distributed database known for high write scale handle multi-key consistency safely. The comments were far less excited about the feature than about Cassandra’s long-running operational pain, with experienced operators saying most teams should still avoid it unless they truly need its scale model.
-
-
Radiation damage to Hubble has been 4.3 years out of phase with the Solar cycle
-
arxiv.org
- 32 comments
A paper analyzing Hubble’s detector damage says radiation-driven degradation has tracked the solar cycle with a 4.3-year phase shift instead of lining up directly with solar activity. Readers in the comments quickly pointed out that Hubble sits in a tricky radiation environment, and that the lag may be more about orbit physics and shielding effects than a mysterious new signal.
-
-
AI companies destroy physical books – let's scan rare books before it's too late
-
annas-archive.pk
- 2 comments
Anna’s Archive posted a call to scan rare and fragile physical books, arguing that AI data collection is accelerating destructive book disbinding and that preservation should happen before hard-to-find copies disappear. There is almost no discussion here because comments were redirected to an earlier Hacker News thread.
-