Top Stories

An Entire Herculaneum Scroll Has Been Read for the First Time

1328 points · scrollprize.org

The Vesuvius Challenge just hit its biggest milestone yet: a complete carbonized scroll from Herculaneum — buried and charred by the eruption that destroyed Pompeii in 79 AD — has been read end to end without ever being physically unrolled. The breakthrough leans on high-resolution CT scans plus machine-learning models trained to detect the faint ink traces inside the rolled papyrus. HN is electrified because this is one of those rare cases where modern ML delivers something genuinely irreversible-in-a-good-way: 2,000-year-old text that was physically impossible to recover is now legible. It’s also a feel-good counterweight to the usual AI-doom discourse.

Om Malik Has Died

926 points · om.co

Om Malik, the journalist and investor who founded GigaOm and spent decades chronicling the rise of broadband, mobile, and the modern tech industry, has passed away. For a huge swath of HN readers, Malik was required reading during the Web 2.0 era — one of the first writers to treat technology coverage as serious business journalism rather than gadget reviews. The comment thread is full of people sharing how his writing shaped how they think about the industry. A genuine loss for tech journalism.

Apple Raises Prices of MacBooks and iPads as Memory Costs Skyrocket

738 points · reuters.com

Apple is raising prices across its MacBook and iPad lineups, pinning the blame on surging memory costs. The backdrop here is the AI-driven DRAM and HBM crunch — hyperscalers buying up memory for AI accelerators have squeezed supply and pushed prices up industry-wide. HN’s interest is twofold: it’s a concrete sign that the AI buildout is now bleeding into consumer hardware pricing, and it reignites the perennial debate about Apple’s storage and RAM margins. When even Apple says it’s eating cost increases, the memory market is genuinely tight.

The ‘Papers, Please’ Era of the Internet Will Decimate Your Privacy

747 points · expression.fire.org

A pointed essay on the global wave of age-verification and identity-check mandates spreading across the web, and what it means to hand over government ID just to access ordinary sites. The author argues these laws — well-intentioned or not — normalize a surveillance architecture that’s very hard to dismantle once built. This hits a deep nerve on HN, where privacy, anonymity, and skepticism of mandatory ID systems run strong. The thread weighs the child-safety motivations against the chilling effects and the inevitable data-breach risk of identity checkpoints everywhere.

Hey Nico, You Didn’t Vibe Code Your Data Room — You Stole It from Papermark

367 points · twitter.com

A very public startup callout: Papermark’s founder accuses another founder of lifting Papermark’s open-source data-room product wholesale and passing it off as something “vibe coded” from scratch. The drama touches on every tense fault line of the current moment — open-source licensing, the AI-coding hype cycle, and what counts as building versus copying when a model can scaffold a clone in an afternoon. HN loves a founder fight, but underneath the spectacle is a real question about attribution and license compliance in the age of AI-assisted development.

IBM Debuts World’s First Sub-1 Nanometer Chip Technology

326 points · newsroom.ibm.com

IBM is touting a research milestone in sub-1nm-class transistor technology, pushing past the limits everyone assumed would slow Moore’s Law to a crawl. As always with IBM Research announcements, HN’s job is to separate the genuine physics from the marketing — “sub-1nm” is a node naming convention, not a literal feature size — but the underlying device innovations are real and feed directly into the next decade of high-performance and AI silicon. The comments dig into gate-all-around transistors, manufacturability, and how long it takes lab breakthroughs to reach fabs.

Apple to Skip High-End M6 Mac Chips in Favor of AI-Focused M7 Line

276 points · bloomberg.com

Apple is reportedly skipping the high-end M6 generation entirely and jumping straight to an AI-focused M7 Pro/Max/Ultra lineup. The strategic read: Apple wants to reset its premium silicon roadmap around on-device AI rather than iterate on the existing design. HN is parsing what “AI-focused” actually means at the silicon level — bigger neural engines, more memory bandwidth for local LLM inference — and whether it signals Apple finally getting serious about running large models on the desktop. Paired with today’s MacBook price hike, it’s a busy day for Apple silicon watchers.

Show HN: OpenKnowledge – Open Source AI-First Alternative to Obsidian/Notion

288 points · github.com

A new open-source, AI-native knowledge base pitched as an alternative to Obsidian and Notion, with retrieval and LLM features baked in rather than bolted on. The “AI-first” framing draws the usual HN crowd that’s wary of yet another note-taking app, but the open-source angle and local-first promises earn it a fair hearing. Discussion centers on data ownership, whether AI features actually improve note-taking or just add noise, and how it stacks up against the entrenched incumbents people have years of muscle memory in.

Zig’s New bitCast Semantics and LLVM Back End Improvements

242 points · ziglang.org

The latest Zig devlog covers reworked @bitCast semantics and meaningful improvements to the LLVM back end. For the systems-programming contingent on HN, Zig remains the most-watched C alternative, and these incremental-but-substantive changes — tighter cast semantics, better codegen — are exactly the kind of careful language design the community follows closely. The thread gets into the tradeoffs of Zig’s self-hosted compiler versus LLVM and where the language is heading as it marches toward stability.