Today’s Highlights#

The best place to start today is a long investigation on libroot.org that reconstructs the life and sudden silence of the Snowden archive. It stitches together six years of publication and abrupt stops across newsrooms to ask an uncomfortable question: what happens when the largest trove of evidence about government mass surveillance still has much unpublished and slips quietly out of sight? That thread of the visible and the invisible runs through the rest of the issue. A small app called ZuckOff tries to make hidden cameras half-visible again, while Attention is all you have shows how recommendation systems quietly take over what we see. Grok 4.7 and Kev point to AI getting better at long tasks and structured judgment, while a change to Disney+ terms and a reflection on Sun Microsystems test whether promises survive execution. The Effect of CRTs on Pixel Art closes with a clear-eyed look at nostalgia versus display technology.

Policy and Governance#

Why the Snowden archive went silent after 2019#

A long investigation on libroot.org traces the Snowden archive from distribution to dormancy. In 2013 Snowden shared an encrypted archive called Pandora, described as about 50,000 documents, with several outlets. The Guardian, which later said it held around 58,000 documents, published only about 30 before stopping in February 2014; The Washington Post, The New York Times, and Der Spiegel also stopped between 2014 and 2015. The Intercept kept publishing until it shut down its archive in March 2019 and released a final batch in May 2019, after which, the piece reports, no outlet has published a new Snowden document. The author contrasts explanations that cited aging material, limited resources, or fading interest with evidence that much material on wartime intelligence and other topics was never released, and that holders of full copies have not publicly explained the freeze or what limits, if any, still apply.

The digest’s reading is that this is less about news cycles and more about editorial independence and public interest. Top comments argued the Guardian’s close consultation with the British government under the D-Notice system — a voluntary arrangement where outlets seek advice before publishing national-security material — blurred the line between vetting and self-censorship. Others in the thread saw a broader attention shift in a polarized media environment, while several readers pushed for academic custody of the archive so its fate does not rest with a handful of journalists and funders.

Discussion: Hacker News thread

Tech and Products#

ZuckOff makes camera glasses audible over Bluetooth#

ZuckOff, built by Polish developer Pawel Szydlowski, turns “is there a camera in the room?” into a check you can run on your phone. According to the project site, it listens for Bluetooth broadcasts — manufacturer identifiers and service UUIDs — to flag nearby camera glasses such as Ray-Ban Meta, Oakley Meta, and Snap Spectacles, showing evidence for each detection, logging sightings, estimating rough distance from signal strength, and exporting logs as CSV, all on-device without an account. The site also notes limits: some models stay silent, signal strength gives distance not direction, and a detection is not proof of recording. Szydlowski built the fingerprints by buying hardware and now invites captures from additional models.

The community reaction was largely appreciative but cautious. Top comments called it a useful way to make an invisible risk half-visible, while also stressing it is no substitute for venue rules and social norms. Several readers compared it with the open-source Nearby Glasses app and debated the trust trade-offs of closed-source Bluetooth scanners, permissions, and store rankings.

Discussion: Hacker News thread

Grok 4.7 aims at staying steady over long tasks#

xAI’s Grok 4.7 pitches itself as the company’s strongest model for coding and knowledge work, with an emphasis on working longer, checking its own work, and keeping safeguards tight. The post says it uses a larger base model and a longer reinforcement learning run weighted toward problems that take hours, with better verification and long-context handling and native support for the Grok Bot harness. On the benchmarks it lists — CursorBench 4.0, DeepSWE, EEBench, and multi-hour office and terminal suites — it shows gains over Grok 4.6 and competitive results against other frontier models. It is priced at $2 per million input tokens and $6 per million output tokens, and is available in Cursor, Grok Build, the xAI API, and third-party routers, with a faster variant at twice the price.

The digest sees the story as a shift from headline scores to staying correct over hours-long workflows, which matters more for real professional use. The thread’s main split was whether frontier labs are over-indexing on superintelligence benchmarks while users want reliability and control. Several readers also weighed price against practicality, noting that cheaper, good-enough models from China now cover many everyday tasks.

Discussion: Hacker News thread

Kev turns judgment calls into small, measurable models#

Kev is a family of small decision models built on Qwen3.5 and inspired by the Jev architecture, offered in 0.8B, 4B, and 9B sizes. The project says you can ask yes/no, multiple-choice, and rating questions in one request, with each question sharing the same input text but isolated from the others, and you get probabilities back instead of a single label. It is designed to be trained and run locally, speaks the TypeSafe System One API, runs the 4B and 9B models on a 32GB Mac in bf16, and includes a web playground and hosted demos to test option-order effects and a chess example.

Readers liked the probability-first design, which fits the messy reality of tickets that touch returns, shipping, and billing at once. Top comments also focused on practicalities of local inference — latency, confidence calibration, and where to set thresholds before handing off to a human.

Discussion: Hacker News thread

Business and Platforms#

What Sun got wrong was not strategy but follow-through#

In What Sun got wrong, Bryan Cantrill responds to nostalgia around Sun Microsystems, the workstation and server maker that defined much of 1990s and 2000s enterprise computing. His distilled answer: Sun grew bored with the mechanics of running a business. He recounts a 2005 episode where a fast-growing startup running OpenSolaris, Sun’s open-source operating system, wanted to buy a lot of Sun hardware — exactly the outcome open-sourcing was meant to create — but could not get Sun to return calls, and when it did, got pitched the wrong product. By contrast, Dell replied overnight, worked the leasing and pricing tier, and did most of the paperwork. Cantrill argues that operational discipline matters as much as strategy, and that no vision survives neglect of sales, fulfillment, and support.

Commenters largely agreed, many adding first-hand memories of slow channels and misaligned incentives at Sun. The thread’s takeaway was to honor what Sun got right in open engineering culture while studying what it got wrong in execution.

Discussion: Hacker News thread

How Disney+ rewrote ad-free to mean almost ad-free#

A Consumer Rights Wiki page on the Disney+ ad policy change, summarized by Heise, describes a January 2025 update to the Disney+ Subscriber Agreement. Section 2(k) recasts tiers described as “no ads” or “ad-free” as “generally free of commercial interruptions, with certain exceptions,” covering cases tied to streaming rights, live or linear content and special events, plus bundle promos, branded integrations, and sponsorship messages, across all tiers including Premium. The wiki notes that subscribers who do not agree are bound unless they cancel. A September 2026 email to German subscribers, framed as a clarification, stated that ads could appear before and after content even on Standard and Premium.

HN readers focused on the contract and trust issue. Top comments argued that burying material changes in lengthy terms and treating continued use as consent empties the promise of a premium tier. Several readers compared similar moves at other streamers and suggested that only regulatory scrutiny or class actions may check a shift from ad-free as a choice to ads as the default.

Discussion: Hacker News thread

Society and Culture#

Attention is all you have#

Alice GG’s essay Attention is all you have starts from the Tetris effect — the way what you stare at rewires what you notice — to argue that most of our attention today is not chosen but assigned. The piece says you open YouTube for a cooking video and get war and market doom, open Spotify and get AI-generated filler between real songs, open LinkedIn or Reddit and get takes tuned for engagement. It contrasts this with an earlier, intentional web of bookmarks and RSS feeds where you decided what to visit. The author suggests reviving that slower web: blogs, tutorials, and feeds that update rarely but on your terms, accepting that less is the point.

The thread resonated strongly. Many readers shared stories of deleting feeds and rediscovering books. Others noted the irony that Hacker News itself can be a refresh trap. A separate strand debated why bookmarks died — link rot and maintenance costs pushing people toward search — and whether a modern, automated full-text history could bring them back.

Discussion: Hacker News thread

CRTs and pixel art: fuzz helped, but not how you remember#

The Effect of CRTs on Pixel Art, a Summer 2024 essay, pushes back on the viral claim that old pixel art was designed to look right only through CRT fuzz. The author says that is half true. Using examples from Commodore 64’s Maniac Mansion and Mayhem in Monsterland and Amiga’s Silkworm, the piece shows how dithering — a pattern of dots that simulates new colors — and anti-aliasing — smoothing jagged edges with intermediate shades — already simulated higher detail within tight color and memory limits, and how CRT scanlines, color bleed, and signal quality could soften them further. But results varied widely by connection: antenna RF and composite video were fuzzy, S-Video was better, and RGB SCART or VGA monitors were quite sharp, sometimes as crisp as modern flat panels. The takeaway is that the soft glow people remember came mostly from cheap consumer TVs, not CRT technology itself.

Readers added lived detail: those who used professional monitors such as the Commodore 1084S remembered clear pixel edges, while those who watched via aerial on the family TV recalled blur. Several commenters argued modern retro games that lean into chunky blocks misread history, and that faithfully revisiting the past means weighing signal, dot pitch, and aspect ratio together.

Discussion: Hacker News thread

Closing#

From the quiet after a major leak to the quiet before a video starts, from feeds that decide for us to hardware that once failed to pick up the phone, today’s through line is about legibility. Can we see what is being decided about us, and can we choose again? The Sun story is a reminder that vision without follow-through fails. The attention essay offers a personal counter-move: rebuild a small, intentional web you control, and give slower media time to teach you again. See you next time.