Apr 25, 2026

Hacker News Morning Brief: 2026-04-25

Here are today’s 30 freshest stories from Hacker News.

Business & Industry

Google plans to invest up to $40B in Anthropic(https://www.bloomberg.com/news/articles/2026-04-24/google-plans-to-invest-up-to-40-billion-in-anthropic)

Google will invest $10 billion in Anthropic at a $350 billion valuation, with another $30 billion contingent on performance targets. The deal strengthens ties between two companies that are simultaneously partners and rivals in the frontier AI race. Google commits the initial $10B now in cash; further funding supports significant expansion of Anthropic’s computing capacity.

HN Discussion: Commenters flagged concerns about “circular” deals—Anthropic previously signed a separate agreement to buy gigawatts of next-generation TPU capacity from Google and Broadcom. Several noted that frontier model makers are becoming commoditized while still needing every major tech company as an “insurance policy.”

Do I belong in tech anymore?(https://ky.fyi/posts/ai-burnout)

A design engineer recounts quitting a well-compensated remote position maintaining our design system—where they doubled component surface coverage and improved accessibility—because generative AI was creating a “psychic toll” on meaningful work. The author describes scenarios including AI meeting notes mischaracterizing discussions, Slack chatbots generating unchecked responses, and 12,000 lines of authentication code merged after AI-agent reviews with no human reading the changes.

HN Discussion: Commenters shared similar burnout experiences from seeing AI-generated code reviewed by other AI agents without oversight. Several debated whether this signals a fundamental misalignment in how companies adopt AI or simply exposes pre-existing problems with code review quality.

Diatec, known for its mechanical keyboard brand FILCO, has ceased operations(https://gigazine.net/gsc_news/en/20260424-filco-diatec/)

Japanese keyboard manufacturer Diatec, creator of the FILCO brand known for high-quality mechanical keyboards using genuine Cherry MX switches, has ceased operations. The news resonated with enthusiasts who viewed FILCO as one of the last brands making keyboards without software configuration requirements—simply plug in and use, with no driver installation needed for basic functionality.

HN Discussion: With 38 comments, readers mourned the loss of a brand that prioritized build quality over RGB customization and macro programming. Several discussed alternatives like Leopold and Keychron while questioning whether the mechanical keyboard market has become too fragmented for any single brand to remain competitive.

The mail sent to a video game publisher(https://www.gamefile.news/p/panic-mail-arco-despelote-time-flies-thank-goodness-teeth)

Since mid-2024, Panic has been receiving extraordinary volumes of fan mail in their Portland office—transforming part of their workspace into what their head of marketing called a “Christmas mailroom.” Senders included a dead fly (intentional), a wedding invitation, a peanut butter cookie recipe, an iPod Nano with custom plaque, needlepoint crafts of Arco characters, and a child’s tooth (unintentional). The letters stem from Panic’s customer rewards program across their Playdate system and games like Thank Goodness You’re Here, Arco, and Despelote.

HN Discussion: Commenters found the story heartwarming but raised questions about whether it reflects a healthy creator-audience relationship. Several noted that Panic’s small size and indie ethos make them uniquely positioned to handle such volume personally, something a larger publisher would almost certainly automate away.

History & Science

Iliad fragment found in Roman-era mummy(https://phys.org/news/2026-04-archaeological-mission-oxyrhynchus-homer-iliad.html)

A papyrus fragment containing the catalogue of ships from Book II of Homer’s Iliad was discovered inside a Roman-era mummy at Oxyrhynchus (modern Al Bahnasa), Egypt, dating to approximately 1,600 years ago. The University of Barcelona team found it deliberately placed in the mummification process—a first for a Greek literary text in such a funerary context. Previously found papyri in similar positions contained only magical or ritualistic content.

HN Discussion: Commenters were fascinated by the cultural implications—why use a secular epic as embalming material? Several drew parallels to Roman practices of recycling Greek texts, while others noted that the physical papyrus’s durability through 16 centuries is remarkable compared to parchment survival in Egypt’s dry climate.

Turbo Vision 2.0 – a modern port(https://github.com/magiblot/tvision)

Magiblot’s project is a cross-platform Unicode-aware port of the classic Borland Turbo Vision framework, originally written for DOS terminal UIs in the early 1990s. Starting as a personal project in late 2018, it reached feature parity with the original by mid-2020 and now compiles on Linux, Windows (MSVC/MinGW/Borland C++), and via Vcpkg, with full source-level compatibility for legacy Turbo Vision applications.

HN Discussion: Commenters appreciated the preservation effort—some recalled writing actual applications with Turbo Vision—and questioned whether modern alternatives like ncurses or ratatui offer cleaner paths today. Several suggested Unicode integration is the killer feature making this genuinely useful for legacy porting.

Paraloid B-72(https://en.wikipedia.org/wiki/Paraloid_B-72)

Paraloid B-72 is a thermoplastic acrylic resin (ethyl methacrylate–methyl acrylate copolymer) originally developed by Rohm and Haas as a surface coating for flexographic ink. It later became the adhesive of choice among conservator-restorers for ceramic and glass conservation, fossil preparation, piano hammer hardening, and museum object labeling. B-72 is valued for being stronger and harder than PVA without being brittle, while remaining more flexible than many competing adhesives.

HN Discussion: Comments focused on the unusual career trajectory of a materials science compound—industrial coating → museum conservation standard—and whether modern synthetic consolidants outperform traditional animal glue in reversibility tests. Several conservators shared experiences with its acetone solubility and long-term aging characteristics.

Humpback whales are forming super-groups(https://www.bbc.com/future/article/20260416-the-humpback-super-groups-swarming-the-seas)

Photographers Monique and Chris Fallows documented 304 individual humpback whales in one day off the west coast of South Africa in December 2025—the greatest number of large whales ever identified in a single day. The super-groups, photographed over two days (208 whales on Dec 29 and 304 on Dec 30), represent a remarkable recovery from industrial whaling that left less than 5% of pre-whaling populations. Southern hemisphere humpback numbers have increased up to 12% per year since the 1986 global moratorium.

HN Discussion: Commenters discussed population dynamics and whether unusual ocean temperatures or prey concentrations are driving the aggregation behavior, as opposed to simple recovery from near-extinction. Some raised concerns about entanglement risks with increased whale densities near shipping lanes and fishing grounds.

ENIAC’s Architects Wove Stories Through Computing(https://spectrum.ieee.org/eniac-80th-anniversary-weaving)

Marking ENIAC’s 80th anniversary, IEEE Spectrum explores how co-inventor John Mauchly and one of its six original programmers, Kathleen “Kay” McNulty, approached computing through a shared Irish word—ríomh, meaning both to compute and to weave, narrate, or compose poetry. The article follows Mauchly and McNulty’s granddaughter Naomi Most reflecting on her grandparents’ dual commitment to prediction (weather forecasting) and storytelling.

HN Discussion: Commenters praised the humanistic angle on computing history—several noted that early programmers were predominantly women whose contributions remain undercredited—and debated whether ríomh’s semantic overlap between weaving and computing is a linguistic coincidence or reflects shared patterns of structured information flow.

The Classic American Diner(https://blogs.loc.gov/picturethis/2026/04/the-classic-american-diner/)

A Library of Congress photo essay documents the American diner through its distinctive aluminum-clad, train-car form factor—mass-produced in the 20th century and shipped inside actual rail cars for delivery. Images span from a 1940 Maryland diner selling hot dogs for 5¢ to a 1959 NYC breakfast platter at 75¢, showing how corrugated metal exteriors and drive-through service windows became iconic elements of American roadside architecture.

HN Discussion: Commenters shared personal memories of local diners, debated the environmental impact of aluminum construction versus steel, and discussed how diner culture has persisted in some regions while being replaced by chain restaurants elsewhere.

Academic & Research

There Will Be a Scientific Theory of Deep Learning(https://arxiv.org/abs/2604.21691)

A paper by Simon, Kunin, Atanasov, Boix-Adserà, Bordelon, and others makes the case that a scientific theory of deep learning is emerging across five growing bodies of work: (a) solvable idealized settings providing intuition for realistic training dynamics, (b) tractable limits revealing fundamental learning phenomena, (c) simple mathematical laws capturing macroscopic observables, (d) hyperparameter theories disentangling them from the rest of training, and (e) universal behaviors shared across systems.

HN Discussion: With 75 comments, the thread was notably technical—some theorists expressed caution about premature unification while others welcomed the synthesis. Several identified specific gaps (generalization bounds, finite-width effects) that remain unresolved, and there was debate over whether the neural tangent kernel framework is sufficient or whether new mathematical tools are needed for modern architectures.

Cosmology with Geometry Nodes(https://www.blender.org/user-stories/cosmology-with-geometry-nodes/)

MohammadHossein Jamshidi, a physics PhD student at Shahid Beheshti University, uses Blender’s geometry nodes to visualize cosmic microwave background (CMB) radiation for cosmology research. The CMB reaches us from the early universe at about 2.7 Kelvin with temperature fluctuations of $10^{-6}$ to $10^{-4}$ K across the sky. All Blender files are available freely on GitHub under CosmicBlenderNodes, demonstrating how open-source 3D software can serve scientific visualization needs.

HN Discussion: Commenters appreciated the cross-disciplinary approach—several in astronomy or plasma physics noted that Blender has become a surprisingly capable visualization tool for research beyond entertainment. Some asked about reproducibility (whether different geometry node implementations produce consistent visualizations of the same data) and standards for scientific visualization using procedural node systems.

A disabled kea parrot is the alpha male of his circus(https://www.cell.com/current-biology/fulltext/S0960-9822(26)00259-9)

Bruce, a wild-born kea (Nestor notabilis) missing the upper part of his bill, has become the alpha male of his circus troupe through self-care tooling innovation. Published in Current Biology, the research documents how the parrot developed novel tool-use behaviors to compensate for his physical limitation—tooling being a complex embodied form of tool use that transforms animal and object into a single problem-solving system.

HN Discussion: Commenters were fascinated by the intersection of disability adaptation and cognitive science, noting that keas (a New Zealand parrot in the corvid family) are already known for complex cognition. Some pointed out the irony that circus environments—which remove animals from natural habitat—became contexts where adaptive innovation was rewarded.

Security & Privacy

Replace IBM Quantum backend with `/dev/urandom`(https://github.com/yuvadm/quantumslop/blob/25ad2e76ae58baa96f6219742459407db9dd17f5/URANDOM_DEMO.md)

A 29-line patch to a Q-Day Prize submission replaces the IBM Quantum backend in solve_ecdlp() with os.urandom, generating uniform-random bitstrings of identical length. The circuit construction, ripple-carry oracle, extraction pipeline, and verifier all run byte-for-byte unchanged—yet every recovered private key matches the author’s reported hardware result exactly across 4-bit to 10-bit challenges.

HN Discussion: Commenters concluded this either exposes a bug in the quantum algorithm’s implementation, reveals that “quantum advantage” was an artifact of measurement or post-processing, or demonstrates that classical random guessing can replicate results when circuit depth is too shallow for meaningful quantum signals.

My audio interface has SSH enabled by default(https://hhh.hn/rodecaster-duo-fw/)

A Rodecaster Duo podcast interface shipped with SSH enabled by default using hardcoded public keys and no signature checks on incoming firmware. The author found the firmware as a simple gzipped tarball, identified two partition-based failover logic, and discovered the device accepts any key matching the embedded pubkey—allowing anyone on the same network to gain root access without authentication changes.

HN Discussion: Commenters noted this is not unusual for IoT devices but surprising at this price point (~£600) from a professional audio brand. Several discussed the broader issue of consumer electronics with exposed management interfaces, and whether embedded Linux distros like Buildroot or Yocto could improve default security postures.

You don’t want long-lived keys(https://argemma.com/blog/long-lived-keys/)

The article argues that long-lived static keys compound risk over time as people leave organizations, cryptographic usage limits degrade security guarantees, and leaked credentials accumulate undetected. It advocates replacing them with ephemeral keys (1-day validity or less), citing examples: EC2 Instance Connect instead of static SSH keys, GitHub trusted publishers for PyPI tokens instead of static tokens found in personal password managers, and short-lived IdP assertions replacing long-lived passwords in SSO.

HN Discussion: Commenters agreed but pushed back on completeness—noting that even SSO requires long-lived signing keys at the IdP level. Several shared horror stories of forgotten SSH keys surviving employee departures for years, and discussed mTLS as a complementary approach where short-lived certificates are managed by tools like cert-manager or Vault.

Web & Infrastructure

”Plain text has been around for decades and it’s here to stay.” – Unsung(https://unsung.aresluna.org/plain-text-has-been-around-for-decades-and-its-here-to-stay/)

The author traces a resurgence of ASCII/plain-text diagramming and UI tools (Mockdown, Wiretext, Monodraw) as constrained interfaces suited for low-key diagramming in source code, increasingly used as an entry point to generative AI. The piece connects these to 1970s–1980s TUIs like Turbo Vision but updated with modern affordances—web access, mouse support—and argues that constraint practice will become more valuable as computers grow more capable and AI requires deliberate restriction.

HN Discussion: Commenters praised the analysis of monospace text as both portable format and potent editing interface, noting they use ASCII diagrams in code comments because they don’t need rendering engines or IDE plugins. One commenter argued that plain-text tools are the most “AI-native” interface paradigm since LLMs output text natively.

Email could have been X.400 times better(https://buttondown.com/blog/x400-vs-smtp-email)

The article traces how the 1984 X.400 standard for Interpersonal Messaging included features that SMTP didn’t get until decades later: message recall and supersede, scheduled delivery, auto-destruct on unread, email threading linking messages like a knowledge base, organization-wide messaging with guaranteed delivery, non-ASCII multilingual support (8 years before MIME), read receipts (15 years early), and built-in encryption. SMTP won because it was simpler to implement—“like a car with no brakes or seatbelts.”

HN Discussion: Commenters were surprised that X.400’s richer feature set lost to SMTP’s simplicity, drawing comparisons to other “better” standards that lost (QWERTZ vs QWERTY, Blu-ray vs HD DVD). Several discussed how modern email clients already re-implement many X.400 features as afterthoughts rather than protocol-level capabilities.

Firefox Has Integrated Brave’s Adblock Engine(https://itsfoss.com/news/firefox-ships-brave-adblock-engine/)

Firefox 149 quietly ships adblock-rust, Brave’s open-source Rust-based content blocking engine, under Bug 2013888 filed by Mozilla engineer Benjamin VanderSloot. The engine supports uBlock Origin-compatible filter syntax, network request blocking, and cosmetic filtering. It is included but disabled by default with no UI or filter lists. Waterfox has already adopted Firefox’s implementation directly.

HN Discussion: Commenters were split—some praised the open-source collaboration between competitor browsers, while others expressed concern about integrating an engine from a company with controversial privacy practices (Brave’s BAT model). Several noted that shipping without enabling it provides no end-user benefit yet.

Work with the garage door up(2024) — (https://notes.andymatuschak.org/Work_with_the_garage_door_up)

Andy Matuschak advocates for “working with your garage door up”—sharing unfinished work, the problems you’re pondering, and how projects don’t work yet. He contrasts this with only announcing finished products, drawing parallels to Screenshot Saturday, Twitch streams, and anti-marketing from Michael Nielsen. Showing process builds more invested followings over time and helps avoid the corruption described in “pitching out corrupts within.”

HN Discussion: Commenters shared examples of creators who successfully built audiences by sharing their working process—some from game development, others from research—and debated the tradeoffs between transparency and the pressure to always appear competent. Several connected it to “building in public” movements in indie hacker circles.

System Administration

Show HN: Lightwhale – a nice home server OS(https://lightwhale.asklandd.dk/)

Lightwhale boots from ISO into a fully functional Docker Engine without installation or configuration steps. Its immutable root filesystem minimizes attack surface and accidental modification—every boot is consistent by design. Customizations and data live on a physically separate storage device, with RAM-based volatile storage as the default and optional persistent partitions that auto-detect on separate hardware.

HN Discussion: Commenters praised the simplicity for home labs but raised questions about managing non-container services (DNS, DHCP) alongside Docker workloads. Several discussed whether immutable Linux distros are practical for users who need frequent system configuration changes versus those wanting “it just works.”

Reverse-engineering infrared-based electronic shelf labels(https://www.furrtek.org/?a=esl)

The author reverse-engineers infrared-based electronic shelf labels (ESLs) used by major retail chains, documenting brands and technologies including MAC layer, PHY, image format, tag electronics, and transmitter hardware. The piece challenges manufacturers’ claims about ecological benefits and notes mischievous attacks are possible: changing prices, locking tags up, draining batteries remotely. A PrecIR project on GitHub provides working code examples with a compatible USB IR interface called ESL Blaster.

HN Discussion: Commenters discussed security implications—many retail chains use ESLs without anti-theft devices—and noted parallels to RFID skimming concerns. One commenter shared work on detecting ESL jamming in their own store’s systems. Several criticized the environmental claims as “utter bullshit” given the resources required to produce tag electronics far outweigh paper label savings.

CC-Canary: Detect early signs of regressions in Claude Code(https://github.com/delta-hq/cc-canary)

CC-Canary is a regression detection system for Claude Code that runs continuously against code modifications to catch quality regressions before they reach production. By monitoring Claude Code’s output across changing inputs, it identifies degradation in coding performance that might otherwise go unnoticed during routine usage or model updates.

HN Discussion: With 24 comments, users debated whether automated regression detection for AI coding tools is necessary or if human code review remains the primary safety net. Several shared experiences with “canary” patterns in ML systems and questioned whether monitoring a coding agent’s outputs produces actionable signals versus noisy false positives from stochastic model behavior.

Other

A 3D Body from Eight Questions – No Photo, No GPU(https://clad.you/blog/posts/questionnaire-mlp/)

The authors describe a system that reconstructs a 3D body model from just 8 questionnaire inputs (height and weight) using a small MLP trained with a physics-aware loss function. It runs in milliseconds on CPU without photos or GPUs, achieving height accuracy of 0.3 cm, mass accuracy of 0.3 kg, and BWH measurements within 3–4 cm—surpassing their previous photo-based pipeline on circumferences while addressing privacy concerns by eliminating the need for user-submitted photographs.

HN Discussion: Commenters appreciated the privacy-first approach over photo-based reconstruction and asked about training data distribution. The author noted that Bartol et al.’s regression from height+weight alone predicts 15 body measurements at 1.2–1.6 cm MAE, but their model also accounted for “muscle weighs more” effects and corrected inconsistencies in the Anny model’s mass calculation.

Education must go beyond the mere production of words(https://www.ncregister.com/commentaries/schnell-repairing-the-ruins)

A Catholic education commentary invokes John Milton’s 1644 Of Education—which defined learning as “to repair the ruins of our first Parents”—arguing that generative AI “industrializes” the pedagogical mistake of demanding finished performances before students undergo reading, questioning, and revision. The author warns against confusing verbal fluency with understanding, noting AI can produce polished prose without genuine engagement with the underlying question.

HN Discussion: Commenters debated whether this is a new problem or just a modern version of longstanding concerns about cheating. Several educators argued that portfolio-based and oral assessment become more valuable precisely because they resist automated generation, while others noted schools are already moving toward process-focused evaluation (drafts, reflections) over product-focused grading.

MacBook Neo and how the iPad should be(https://craigmod.com/essays/ipad_neo/)

A long-form essay arguing against Apple’s approach to making the iPad a Mac replacement, focusing on why touch-on-Mac is ergonomically flawed. The author, with 20 years of Mac use and 10 years with iPads, argues that reaching up from keyboard to touchscreen quickly gets fatiguing and that Apple should embrace touch-first design for tablets rather than bolting desktop paradigms onto a touch interface.

HN Discussion: With 135 comments, the thread was broadly positive on the core argument—several Windows 2-in-1 users confirmed similar ergonomic issues with touchscreen laptops. Some asked why Apple won’t offer a detachable form factor like Microsoft’s Surface instead of trying to merge two experiences in software alone.

DeepSeek V4(https://api-docs.deepseek.com/news/news260424)

DeepSeek has released V4 Preview with open-weights models: V4-Pro (1.6T total / 49B active params) targeting frontier-tier agentic coding, and V4-Flash (284B total / 13B active params) for fast inference. Both support 1M context as the default, use novel token-wise compression with sparse attention (DSA), and are integrated with Claude Code, OpenClaw, and other agent frameworks. Legacy models retire July 2026.

HN Discussion: With over 1,400 comments, the thread centered on parameter efficiency—comparing active vs total parameters favorably against Mixtral-style MoE—but skepticism about whether “world-leading long context” holds up with real documents. Many noted aggressive deprecation of legacy model IDs and questioned positioning against Gemini-3.1-Pro and GPT-5-class models.

Show HN: VT Code – Rust TUI coding agent with multi-provider support(https://github.com/vinhnx/VTCode)

A terminal-based coding agent built in Rust that supports multiple AI providers through a unified TUI interface. The project brings coding agent functionality directly into the terminal without requiring a browser or desktop application, leveraging Rust’s performance characteristics for low-latency interaction.

HN Discussion: Early observers noted the growing category of terminal-based coding tools and questioned which providers are supported, whether it handles file I/O beyond generating code snippets, and how it compares to existing terminal IDE approaches like Helix or Neovim with AI plugins.

Overtom Chess Computer Museum(https://tluif.home.xs4all.nl/chescom/Engindex.html)

A personal museum website cataloguing vintage chess computers from the 1980s and 1990s by manufacturer: CXG/Sphinx (36), Excalibur (31), Fidelity (29), Mephisto (50), Novag (53), Scisys/Saitek (86), Tandy/Radio Shack (23), plus a miscellaneous section with 103 entries. The site serves as both an archive and technical reference for collectors preserving these pre-internet gaming machines.

HN Discussion: Commenters shared nostalgia for their own chess computers—several mentioned specific Mephisto and Novag models—and debated the computational challenges of building dedicated hardware engines before GPUs became accessible for game AI, noting many used custom ASICs and lookup tables for endgame databases given 1980s silicon constraints.

The Oxford All Souls General Examination(2025) pdf

Oxford’s All Souls College released its 2025 General Examination papers—a notoriously competitive postdoctoral fellowship exam. The PDF reveals the breadth and depth of questions expected for one of academia’s most selective appointments, covering social and economic systems in ways that test candidates on interdisciplinary reasoning rather than narrow specialization.

HN Discussion: With 21 comments, readers reacted with a mixture of awe and horror at the scope of knowledge required. Several discussed whether All Souls’ selection process still serves its purpose in modern academia or if the fellowship’s prestige has become self-reinforcing without commensurate intellectual contribution. Some former candidates shared personal experiences navigating the examination’s interdisciplinary demands.