Hacker News Morning Brief: 2026-06-08


A new hepatitis B drug trial demonstrates functional cure rates, Troy Hunt marks 1,000 breaches in Have I Been Pwned while disclosure lag worsens, and a 7.8 magnitude earthquake strikes the southern Philippines. Elsewhere, Firefox merges Vulkan Video support, Stanford researchers expose hiring algorithm monocultures, and Ken Shirriff powers up a vacuum-tube module from 1948.


Academic & Research

New drug ‘functionally cures’ many hepatitis B virus infections

Summary: A new drug trial reports that 19% of treated patients achieved HBsAg clearance, with results independently replicated across more than 1,800 patients. The trial enrolled non-cirrhotic patients with moderate baseline HBsAg (100–3,000 IU/mL) already on stable nucleotide analogue therapy, a selection that matters because HBV-related deaths cluster in patients with higher antigen loads and advanced disease. The drug’s mechanism targets the HBsAg production pathway rather than merely suppressing viral replication, representing a shift from suppression toward functional cure.

HN Discussion: Commenters questioned why HBV cure research persists when a vaccine has existed for 40 years, with some redirecting attention to incurable viruses like HSV and HPV that lack comparable vaccine coverage. One commenter asked whether treated patients remain contagious and whether transmitted virus from treated patients would behave differently in a new host. The trial’s selection bias drew scrutiny: the 19% rate applies to a carefully screened non-cirrhotic population, not the full disease spectrum.

The Smallest Brain You Can Build: A Perceptron in Python

Summary: This tutorial builds a perceptron from scratch in Python with interactive browser demos, covering weights, bias, decision boundaries, epochs, and learning rate using no external libraries. It explains Frank Rosenblatt’s 1958 concept through minimal math, walking through normalization, step functions, and why convergence is guaranteed only for linearly separable data. The author writes for readers who need concepts explained slowly from the ground up rather than through dense notation.

HN Discussion: Commenters praised the interactive demo’s clarity for ML beginners, with comparisons to similar minimalist projects like NanoNeuron in JavaScript. Some debated whether ad-hoc demos effectively teach fundamentals versus structured resources like fast.ai or Bishop’s Deep Learning textbook. The author’s non-native-English pedagogical perspective was appreciated for making the content more accessible.

A visual introduction to kernel functions

Summary: This walkthrough introduces kernel functions using a cheese-to-gold metaphor to explain how data can be mapped into higher-dimensional spaces for linear separation. It covers the kernel trick — computing inner products in transformed feature spaces without explicit transformation — and walks through common kernels (linear, polynomial, RBF) with intuitive visualizations of decision boundaries. The writing targets readers who found standard textbook explanations of kernels opaque.

HN Discussion: Readers praised the readability and thumbnail-style visual explanations. Some confusion arose from the title, with commenters expecting operating system kernels rather than ML concepts. Appreciation came from those who had previously struggled with kernel theory in traditional course materials.


Security & Privacy

1k Data Breaches Later, the Disclosure Lag Is Worse

Summary: Troy Hunt marks the 1,000th data breach loaded into Have I Been Pwned, reflecting on why the service remains necessary 12.5 years after its launch. Disclosure lag times are worsening: in Carnival’s ShinyHunters attack, 8.7 million records were published while formal notification came days later, and 85% of those records were already in HIBP from prior breaches. Despite GDPR and CCPA, companies increasingly delay notification, and many breaches are discovered only after data appears online.

HN Discussion: Commenters questioned whether corporations have genuine business motivation to disclose faster than legally required, with most concluding they do not. One pointed out that Google and Apple now throttle app hotfix reviews to a minimum three-day window, slowing vulnerability patches for mobile applications. Debate emerged over whether HIBP remains necessary versus alternatives that reveal exactly what data leaked rather than merely confirming exposure.

SDSU Wired Its Dorms with 1,300 AI Cameras Without Telling Students

Summary: San Diego State University spent over $1.3 million installing 1,300 AI-enabled cameras across campus — including 330-plus in residence halls — without notifying students, who discovered the full scope through their own student newspaper. University Police completed the installation in 2024 across classroom buildings, bookstores, dining areas, parking structures, gyms, and dorms. The deployment is one of the largest in the California State University system.

HN Discussion: Commenters clarified that cameras appear to be in dorm common areas like halls and lobbies rather than inside individual rooms. Concern centered on the shifting legal standard: “no reasonable expectation of privacy” in public areas is becoming “reasonable expectation of no privacy.” Questions remained about what “AI-enabled” concretely means — facial recognition, behavioral analysis, or simply networked motion detection.


AI & Tech Policy

DeepSeek V4 Pro beats GPT-5.5 Pro on precision

Summary: RuntimeWire claims DeepSeek V4 Pro outperforms GPT-5.5 Pro on instruction-following precision, schema matching, and edge-case handling across four experimental tasks testing instruction adherence, structured output compliance, and clean resolution of corner cases. The article positions DeepSeek as stronger in exactness-focused workflows while conceding GPT-5.5 Pro remains competitive overall. The methodology behind the benchmarks is not disclosed in reproducible detail.

HN Discussion: Commenters dismissed the benchmarks as poorly constructed experiments with no reproducibility, reading like AI-generated clickbait. One tester reported GPT-5.5 Pro exhausting a $100 budget on a vulnerability scanning benchmark while DeepSeek V4 Pro completed it for about $1, sparking practical cost-to-performance comparisons against Claude Code Max plans. Skepticism dominated about declaring model superiority from small test sets without verifiable methodology.

Algorithmic Monocultures in Hiring

Summary: A Stanford-led study of 3.4 million real applicants across 156 employers and 11 sectors — all screened by algorithms from the same vendor — reveals large-scale adverse impact: 25.87% of Black applicants and 14.74% of Asian applicants were directed to positions with adverse impact under Title VII standards. Over 90% of US employers use hiring algorithms from a small number of shared vendors, creating a monoculture that bottlenecks opportunity. Adverse impact was visible only through disaggregated population analysis.

HN Discussion: One commenter noted the study measures application-level outcomes rather than individual outcomes, creating confounds in how demographic groups apply to different roles. Comparisons to RealPage’s anticompetitive landlord monoculture highlighted similar systemic risks. Reports circulated of ATS systems caching candidate scores for 3–12 months, meaning a rejection at one company propagates to others using the same vendor. Some argued monoculture hiring problems predate algorithms.

I design with Claude more than Figma now

Summary: A Jane Street designer describes shifting their primary design tool from Figma to Claude Code, using it for rapid UI iteration on OCaml and Bonsai web components. Claude provides unlimited iteration on unfamiliar technologies without judgment, enabling faster exploration than manual tools. The author acknowledges a downside: designing through an LLM constrains creativity to what the model can produce, risking an iterative rather than fluid mindset when working on genuinely new ideas.

HN Discussion: Commenters worried that business stakeholders will increasingly arrive with AI-generated “ready” solutions, reducing openness to holistic design feedback. One noted Jane Street’s Anthropic investment as potential bias. A recurring theme was that LLMs cannot think beyond iteration, requiring humans to deliberately introduce out-of-the-box direction. The claim of “free, unlimited iteration” drew humor from commenters paying substantial Claude subscription fees.


Tech Tools & Projects

APC–2 – A professional record cutter for producing original playback discs

Summary: Teenage Engineering’s APC-2 is a professional vinyl record cutting system featuring direct-drive precision, variable pitch control, and a stereo feedback cutting head. The 140 kg unit includes integrated vacuum hold-down, swarf removal, temperature-controlled heating, and a built-in power amplifier with RIAA encoding. It supports DAW automation including locked grooves and specialty cuts, and is available exclusively through SUPERSENSE with no public price.

HN Discussion: Commenters questioned the economics: recordcut.com charges $12 per single disc, making ownership hard to justify outside boutique studios. The Monty Python three-sided record with concentric parallel grooves came up as a benchmark for what specialty cutting enables. Appreciation for Teenage Engineering’s analog commitment was balanced against uncertainty about who actually needs in-house vinyl production.

A Matter Wi-Fi Light Bulb in Rust on the Raspberry Pi Pico 2 W

Summary: This project implements a fully Matter-compliant smart bulb on the Raspberry Pi Pico 2 W (RP2350) using the rs-matter stack and embassy async framework in pure Rust. It handles BLE commissioning and Wi-Fi connectivity, provisionable through Apple Home and Google Home without proprietary SDKs. The repository includes examples for GPIO control, Wi-Fi setup, and the complete Matter commissioning flow.

HN Discussion: The author explained the project demonstrates that building compliant Matter devices from scratch in Rust is now feasible. Interest focused on the rs-matter stack’s maturity and whether it extends beyond demos to production use. Discussion of embassy’s async model highlighted its advantages for embedded Rust development over traditional approaches.

A discovery about GCC’s unidirectional rotation algorithm

Summary: Raymond Chen investigates GCC libstdc++‘s random-access iterator rotation algorithm, revealing an unexpected implementation detail in how the standard library handles unidirectional rotation for containers lacking bidirectional traversal. The analysis compares GCC’s approach against the expected algorithmic behavior for the C++ rotate primitive, examining edge cases and performance characteristics. It forms part of The Old New Thing’s ongoing documentation of compiler and standard library quirks.

HN Discussion: A commenter felt the article was padded with hyperbole — “shocking,” “striking,” “disappointing” — for what is ultimately a narrow implementation detail. The clickbait-adjacent framing was noted as atypical for Chen’s usually measured style.

My automated doubt development process

Summary: The author describes a multi-agent AI development workflow built around “automated doubt” — front-loading scrutiny through specialized subagents that audit security, performance, and correctness from different perspectives. Process phases include design validation, implementation review, and synthesis where a coordinator deduplicates overlapping critique. The system emerged from lost trust after allowing LLMs too much autonomy too quickly, and each subagent catches defects a single model instance would miss.

HN Discussion: One commenter built a “project drift” assessor that detects orphaned code and stale design decisions in AI-generated codebases. Criticism targeted spec-driven development for assuming requirements are valid rather than questioning whether features should exist at all. The “skeptical pass” technique — a final agent that prevents overengineering of trivial edge cases — was shared as a practical refinement.

Office-open-xml-viewer: Office XML document viewer that renders to HTML Canvas

Summary: This open-source viewer parses Office Open XML documents (.docx, .xlsx, .pptx) and renders content directly to HTML Canvas in the browser, handling text styling, tables, images, and layout properties. It manages the complex ZIP-plus-XML structure of modern Office formats, targeting developers needing lightweight document preview without Microsoft Office or server-side rendering. The project aims to make OOXML inspection accessible client-side.

HN Discussion: Testing against medium-complexity Word documents revealed significant rendering errors, suggesting limited validation with real-world files. An alternative project — docx-editor by eigenpal — was recommended for handling complicated documents more reliably. Another developer shared their work on ooxml-cli for editing PPTX, DOCX, and XLSX via AI agents for corporate template adaptation.


Web & Infrastructure

Show HN: NoSuggest – Watch YouTube without the recommendation algorithm

Summary: NoSuggest is a web app that displays only YouTube channels users explicitly add, with no recommendations, no autoplay, no notifications, and no Shorts loop. Each device maintains its own independent channel list so phones, laptops, and tablets can have curated selections. A Kids Mode with PIN protection hides search and channel management tabs. The service is free with no account or credit card required, currently showing the latest five videos per channel.

HN Discussion: Users compared NoSuggest favorably to maintaining a self-hosted Invidious instance, which requires ongoing upkeep. Requests included stream support alongside videos and settings to display more than five videos per channel. The DF YouTube browser extension was referenced as a similar approach, though NoSuggest’s no-signup model was specifically appreciated.

How’s Linear so fast? A technical breakdown

Summary: This analysis reverse-engineers Linear’s architecture from public talks and app behavior, identifying a browser-local database that enables instant optimistic mutations synced asynchronously to the server. The sync engine replaces the traditional request-spinner-response loop, while first-load optimization, animation strategies, and deliberate UI design contribute to sub-millisecond perceived latency. The author had no access to Linear’s source code but synthesized the approach from observable behavior.

HN Discussion: Recommendations for Zero and Replicache emerged as frameworks for building similar local-first architectures. A daily Linear user pushed back: search is slow, Pulse is noisy, and the UI is clunky despite the impressive perceived speed. A reverse-engineered Linear sync engine by wzhudev on GitHub was referenced. Debate continued over whether local-first syncing truly solves latency or shifts complexity to conflict resolution.

Firefox Merges Support for Vulkan Video Decoding

Summary: Firefox has merged initial Vulkan Video decoding support, scheduled for the Firefox 153 release in July 2026. Vulkan Video provides cross-platform GPU-accelerated video decoding, reducing dependence on VA-API which lacks universal Linux driver support. NVIDIA Linux users benefit directly — the nvidia-vaapi-driver compatibility layer is no longer necessary — and Arm and embedded graphics drivers that lacked VA-API implementations gain hardware video decoding.

HN Discussion: NVIDIA users on Linux celebrated dropping the VAAPI compatibility toolchain entirely. Questions arose about practical impact on YouTube playback and DRM-protected Netflix content. One commenter asked why Firefox implements first-class video decoding rather than offloading to ffmpeg, while jokes about “compiling shaders” messages appearing on video-heavy sites tempered the enthusiasm.


History & Science

Man-Computer Symbiosis J. C. R. Licklider (1960)

Summary: J.C.R. Licklider’s seminal 1960 paper envisions tight coupling between humans and electronic computers, where machines handle routinizable computation and humans provide goals, hypotheses, and evaluations. The paper anticipates time-sharing, advanced memory organization, new programming languages, and improved I/O as prerequisites, arguing the partnership will outperform humans alone on intellectual operations. The biological symbiosis metaphor — fig tree and fig wasp — frames the relationship as genuinely interdependent.

HN Discussion: Commenters expressed admiration for Licklider’s prescience in anticipating interactive computing decades before it materialized, and noted the paper’s continued relevance to contemporary debates about AI-human collaboration.

Powering up a module from the IBM 604: an electronic calculator from 1948

Summary: Ken Shirriff powers up a pluggable module from the IBM 604 Electronic Calculating Punch, a 1948 vacuum-tube calculator that bridged electromechanical tabulating and electronic computing. The 604 was not a stored-program computer but a programmable calculator using plugboards for operation sequencing, performing arithmetic electronically via thyratron tubes. Shirriff traces the circuit path, identifies tube types, and demonstrates the module operating after decades in storage.

HN Discussion: Context on IBM’s 600 line emerged: the 601 (1931) multiplied mechanically via plugboard, the 602 (1946) could divide, and the 602A remained in production into the early 1970s. A reader with over a dozen original IBM modules and circuit diagrams confirmed they will preserve them after reading Shirriff’s analysis. Praise focused on the footnotes explaining plugboard mechanics with primary documentation.


Business & Industry

Texas grid flags risks as data centers, crypto sites fail voltage tests

Summary: ERCOT has flagged grid stability risks after data centers and cryptocurrency operations in Texas failed voltage support tests. The core concern is that massive concentrated loads which disconnect suddenly during faults can cause generator overspeed and frequency excursions on the 60Hz grid. The failures highlight tension between Texas’s rapid data center expansion and the engineering realities of maintaining voltage and frequency stability under concentrated industrial demand.

HN Discussion: One commenter explained the physics: excess fuel keeps generators spinning after sudden load loss until control systems catch up, risking phase and frequency mismatch with the grid. A proposal for $1/W installed-capacity hookup fees would charge a 10GW data center $8 billion, internalizing grid costs currently spread across ratepayers. Discussion of whether large consumers should maintain artificial load banks for gradual ramp-down rather than instant disconnection followed.

An Ohio Valley 100k-watt FM signal is severed in broad daylight

Summary: Copper thieves severed the transmission line of WDGG(FM), a 100,000-watt station in Boyd County, Kentucky, knocking it off the air. Thieves stole 200–400 feet of large-diameter coaxial cable worth an estimated $1,360–$6,400 in scrap, while repair costs run $70,000–$100,000. Station president Mike Kirtner reported the community rallied around the station during the outage.

HN Discussion: Commenters noted Kentucky has an epidemic of utility cable theft, with phone and power lines cut for scrap multiple times per year even in safe rural areas. Discussion of modern solid-state transmitters with foldback protection that can dead-short in under 100ms, preventing catastrophic damage, followed. Comparisons to Detroit’s copper theft epidemic — where thieves targeted power substations, sometimes fatally — underscored frustration with low severity of infrastructure theft charges.

Tech sell-off widens as South Korea index plunges

Summary: South Korea’s KOSPI dropped sharply alongside a broader global tech sell-off concentrated in semiconductor and AI-exposed stocks. Samsung Electronics and SK Hynix together represent 30–40% of the KOSPI by market cap, making the index highly sensitive to AI capex sentiment. The sell-off tracks the same thesis driving Nvidia and TSMC pullbacks, expressed through a country index levered to memory and chip manufacturing. Timing reflects Asian markets catching up to a US Friday decline.

HN Discussion: One commenter argued the KOSPI functions more like a concentrated chip/HBM ETF than a diversified country index. Debate centered on whether the drop represents a genuine correction or merely Asian markets playing catch-up to a US selloff that occurred after their Friday close. An archive.ph link was shared to bypass the FT paywall.


System Administration

What is the purpose of the lost+found folder in Linux and Unix? (2014)

Summary: The lost+found directory exists on Unix and Linux filesystems as the destination for orphaned files recovered by fsck during filesystem checks. When fsck encounters inodes with allocated data blocks but no directory entry pointing to them, it places them in lost+found. XFS, Btrfs, and ZFS do not use lost+found because they employ different consistency models than the ext family and older non-journaled filesystems. The directory is created at filesystem format time and typically remains empty.

HN Discussion: Veteran sysadmins shared recovery stories of entire directory trees emerging from lost+found after disorderly shutdowns corrupted root inodes. Reference to Eric Foxley’s 1985 book Unix for Super-Users documenting lost+found on page 52 placed the concept historically. Many administrators noted decades of experience without ever finding anything there, thanks to journaled filesystems. One memorable case involved a 1TB NAS that appeared empty after a crash but had everything recovered by fsck into lost+found.

Do we fear the serializable isolation level more than we fear subtle bugs (2024)

Summary: This article argues serializable isolation should be the database default because most developers cannot correctly reason about concurrency anomalies at weaker levels. It cites research identifying 22 concurrency vulnerabilities — 5 caused by weak default isolation and 17 by improper transaction scoping where database accesses span independent API calls. The comparison to memory-safe languages is explicit: perceived performance cost masks deeper correctness tradeoffs, and database concurrency bugs are harder to reproduce and debug than threading bugs.

HN Discussion: The biggest practical problem raised: serializable isolation requires handling serialization exceptions and retrying at commit time, which most frameworks don’t support gracefully, causing production failures under load. The vulnerability paper’s finding that 17 of 22 bugs were scope-based rather than level-based meant transaction boundaries mattered more than isolation level. Commenters stressed that developers must understand their specific database’s concurrency model, since Oracle, Postgres, MySQL, and SQL Server all behave differently at the same nominal isolation level.


Geopolitics & War

7.8 magnitude earthquake shakes part of southern Philippines. Tsunami possible

Summary: A 7.8 magnitude earthquake struck off southern Philippines near Sarangani at 07:37 AM local time on June 8, 2026, at a depth of 33km. PHIVOLCS issued tsunami warnings and advised coastal residents to move to higher ground. The quake was felt across multiple provinces at varying intensities, and New Zealand Civil Defence assessed no tsunami threat to the broader Pacific region.

HN Discussion: A first-person account from a 29th-floor resident described rooms shaking for five-plus minutes, walls cracking, and siding falling during evacuation down stairs. PHIVOLCS’s website was overloaded, making X/Twitter the most reliable real-time source. The quake was felt 600km away on upper floors where building sway caused doors to swing and hanging plants to move. New Zealand Civil Defence confirmed no regional tsunami threat.


Other

Building from zero after addiction, prison, and a felony

Summary: Gavin Ray recounts spending ages 14–16 in maximum-security juvenile prison, receiving a felony at 19, losing nearly everything to addiction, and rebuilding through software and open source. He credits his career to people who took chances on him and to OSS contributions that demonstrated competence despite his record. The post explicitly avoids AI-generated prose, which the author considers “deeply disrespectful” in personal writing. It is written for others with criminal records wondering whether a tech career is possible.

HN Discussion: Commenters shared their own non-traditional paths: bike messengers, freight train riders, and self-taught programmers who found tech through unconventional routes. The job market has shifted — the author landed work his first day out, whereas AI resume filters now create barriers for unconventional candidates. Preston Thorpe, cited as inspiration, has a similar story at pthorpe92.dev. Appreciation for the refusal to use AI in personal narrative was a recurring note.

Dopamine Fracking

Summary: The author coins “dopamine fracking” for the practice of pumping disproportionate resources — money, analytics, crowdsourced optimization — into previously casual activities to extract the most concentrated dopamine hit possible. Like hydraulic fracking, the practice yields intense short-term reward but damages the long-term health and sustainability of whatever it targets, from gaming to social media to benchmark optimization. The concept was previously termed “sloptimization,” but the fracking metaphor better captures the destructive dynamics.

HN Discussion: Commenters connected the concept to Marx’s “all that is solid melts into air” as a pattern of capitalist acceleration, and compared it to “human fracking” from the book Attensity!. Debate over naming: some wanted terms incorporating refinement, supernormal stimuli, and ease of consumption. One critic felt the title alone conveyed the full concept, with the body being a general internet-culture rant.

1worldflag: A blue dot on a transparent background

Summary: The One World Flag features a blue sphere symbolizing Earth on a transparent background that adapts visually to whatever is behind it. Designed as a unifying symbol beyond national flags, the project emphasizes shared planetary identity and has been distributed through a German NGO ambassador program where magazine readers pass flags to each other internationally. The transparent material is intentional, making the flag’s appearance context-dependent.

HN Discussion: Commenters noted the novelty of transparency in flag design and debated whether similar effects could be achieved with natural cloth weaves rather than synthetics. Anticipation of a CGP Grey flag review referenced his well-known video on national flag design principles. Humor dominated: one commenter suggested the flag should depict two people arguing, and another simply requested a color change.

Show HN: I Derived a Pancake

Summary: This systematic investigation of pancake batter chemistry covers acid-base neutralization, CO2 production kinetics, gluten inhibition, and Maillard browning for a 125-gram flour batter, with an interactive stoichiometric calculator that adapts to available ingredients. Leavening ratios are derived from first principles rather than traditional recipe proportions, eliminating the need for buttermilk. The piece covers lacy edge formation, crispness factors, and fat selection across 22 minutes of reading.

HN Discussion: Commenters appreciated the parametric recipe-card approach and wanted to apply it to bread baking with adjustable hydration and salt levels. A 10-hour rest requirement drew criticism for disqualifying the recipe from busy mornings. Requests for ingredient toggles — such as replacing pork fat with avocado oil — reflected dietary practicality concerns. One commenter praised the depth but complained about not actually measuring crispness despite having the equipment.

Making peace with your unlived dreams (2023)

Summary: The author reflects on accepting that physical limitations (bad knees) permanently foreclose dreams like snowboarding, framing the psychological process of making peace with irretrievable aspirations. The argument is that life is simultaneously too big and too short to pursue everything, making triage of dreams a necessary skill. A distinction is drawn between genuine personal desires and culturally imposed aspirations amplified by mass media, and grief over unlived lives is treated as legitimate but ultimately something to process.

HN Discussion: Commenters shared deeply personal stories: children with severe illness and autism, career-ending injuries, body composition realities that foreclose activities. The distinction between personal dreams and cultural expectations sold through self-help narratives was explored, with reference to a Polish article on millennials trapped in constant self-development. The “not-knowing” of what might have been emerged as more haunting than accepting a definite limitation.

Splash Is a Colour Format

Summary: Splash is a decimal-based colour format using three digits (0–9 per channel) for red, green, and blue, yielding exactly 1,000 possible colours. The constraint is intentional — designed to prevent decision paralysis when picking palettes — trading resolution for cognitive simplicity. The author uses it for personal theming, acknowledging imperfection but finding the restriction liberating, and the page lists all 1,000 colours exhaustively.

HN Discussion: Criticism focused on insufficient flexibility in neutrals (blacks, greys, whites) combined with inadequate constraint to actually eliminate fussing. Comparison to 3-digit hex colours (#fff) made the decimal format seem like a resolution loss. Appreciation came from viewing it as art rather than spec, with potential for terminal colour schemes. A paint-mixing model — expressing colours as ratios of arbitrary base colours rather than channel values — was proposed as an alternative.