Monday, September 28, 2026 · No. 45
25 articles · ~5h 33m read · 8 sections
A personal morning paper, assembled daily; the selection is the reader's, the words are the authors'.
Sign in to read every article's full text and download the editions, or request access.
Top Stories
Reverse Engineering the iPod Classic's Undocumented Mikey Chip
terminalbytes.com via dpedu · Lobsters · terminalbytes.com · 13 min
Hardware · First personiPod Classic · Rockbox driver · I2C reverse engineering
Matched interests: Hardware HackingMechanical KeyboardsApple
Adding inline earphone remote support to Rockbox on classic iPods requires decoding Apple's mystery I2C jack controller. Sweeping all 256 register values on-device reveals that mode 0x2f captures analog voltage droops from the volume buttons rather than a suspected serial protocol. The resulting driver compensates for a sleepy hardware event engine and delayed release events, submitting 318 lines upstream to restore three-button playback.
Hacker News: 4 points · 0 commentsLobsters: 11 points · 2 commentsComments
Malleable software: Restoring user agency in a world of locked-down apps (2025)
evakhoury · Hacker News: Front Page · inkandswitch.com · 45 min
Software engineering · Analysismalleable software · user agency · end-user programming
Matched interests: Mobile UXZed EditorDeveloper Tools
Physical workspaces naturally adapt to their occupants, yet computerized tools trap workers in static workflows that drive professional burnout, as seen in inflexible electronic health records. Standard remedies like plugin APIs, configuration toggles, and open-source codebases fail because they impose steep technical friction and vendor-gated boundaries. True computing elasticity requires environments that non-programmers can directly modify in the moment to fit bespoke daily workflows.
EX-ARRR: Sailing the 0-click Seas
ironpeak.be via fro · Lobsters · ironpeak.be · 23 min
Software engineering · First personzero-click exploit · OpenEXR parser fuzzing · Apple ImageIO heap overflow
Matched interests: Zero-CopyFuzzingApple
A channel-count mismatch in Apple's libAppleEXR.dylib decoder triggers a heap buffer overflow by allocating twelve bytes per pixel for RGB images while a SIMD routine writes sixteen. Found using disassembly-guided fuzzing, the vulnerability allows roughly 800 kilobytes of attacker-controlled pixel data to overwrite adjacent memory, enabling zero-click exploitation inside privileged daemons handling incoming iMessage payloads.
Made by Mechanical Means
cocoflunchy · Hacker News: Front Page · felixrieseberg.com · 14 min
AI & ML · First personAI-assisted web design · Blender procedural rendering · multi-agent orchestration
Matched interests: ClaudeCreative CodingCreative Automation
Orchestrating roughly sixty parallel Claude threads from a smartphone enables the complete redesign of an interactive homepage without writing local code or touching GitHub. Controlling Blender as a Python library, the setup constructs a 1990s journalist’s room with procedural textures and baked Cycles lightmaps. It even tracks handheld camera motion to render a 2,787-frame CGI parody of a Werner Herzog documentary.
I used an AI detector for a month. What I found from my peers, dates, and heroes horrified me.
Aki Ito · All Content from Business Insider · businessinsider.com · 16 min
AI & ML · First personAI text detection · AI slop · Pangram
Matched interests: AI CriticismCreative WritingText Editors
Auditing digital correspondence with the Pangram classifier reveals synthetic text reaching deep into everyday life, from corporate layoff sympathy to an emergency-room nurse's breakup message. While startup founders justify automated drafts as carrying their authentic intent, analyzing personal archives demonstrates how readily machine phrasing slips into an individual's own voice.
Tech & Engineering
What makes Lisp difficult to read?
paultm.nl via veqq · Lobsters · paultm.nl · 14 min
Software engineering · AnalysisLisp syntax readability · cognitive science of perception · notation design
Matched interests: Functional ProgrammingParser DesignData Structures
Lisp’s reputation for unreadability stems not from mere unfamiliarity, but from syntactic structures that clash with human visual perception. By placing identifiers inside parentheses, Lisp pushes matching delimiters beyond the eye’s two-degree focal region, defeating Gestalt grouping principles like closure and symmetry. Furthermore, uniform whitespace deprives peripheral vision of the variable clustering that C-style notation naturally provides.
Hacker News: 2 points · 1 commentsLobsters: 12 points · 5 commentsComments
Syncing Rust GCC backend or how to test Murphy's law
blog.guillaume-gomez.fr via fanf · Lobsters · blog.guillaume-gomez.fr · 7 min
Software engineering · First persongit subtree sync · Rust GCC backend · CI infrastructure failures
Matched interests: RustGitCompilers
Synchronizing the rustc_codegen_gcc repository with the main Rust compiler turns into a two-month ordeal of cascading build and toolchain failures. Relying on a patched git subtree workflow exposes false-positive CI caching, linker incompatibilities that disrupt Miri, and bootstrap sysroot breakages. The grueling series of reverts, testsuite patches, and tool updates ultimately prompts plans to abandon subtree synchronization for josh.
How Differential Testing Exposed Type and Performance Bugs in Chuks v0.1.2
Tech_World · HackerNoon · hackernoon.com · 9 min
Software engineering · First persondifferential testing · compiler type bugs · bytecode VM
Matched interests: TypeScriptRustFuzzing
Reconciling Chuks' bytecode VM with its native binary reveals that twelve standard array methods silently erase return types through reflective boxing. Old test suites masked the generic interfaces by only comparing printed output, hiding defects until assertions consumed declared types directly. Compiling specialized loops over concrete arrays eliminates the boxing, restoring type safety while dropping 300,000-element execution times from 40ms to under 1ms.
Writing Efficient C++ Code
asawicki.info via jeaye · Lobsters · asawicki.info · 21 min
Software engineering · Analysisdata-oriented design · cache efficiency · C++ performance
Matched interests: Data StructuresMemory AllocatorsQuery Optimization
Excessive object-oriented abstractions and scattered pointer-linked objects degrade performance by triggering cache misses and complicating thread parallelization. Because fetching data from main memory takes around 250 CPU cycles compared to just three for L1 cache, software performance depends heavily on data layout. Structuring programs around contiguous memory arrays and packed data types ensures sequential cache hits while avoiding pointer-chasing heap overhead.
Hacker News: 155 points · 118 commentsLobsters: 4 points · 7 commentsComments
Science & Space
Cognitive Sanctuaries or: Manifesto of the Department Chair
Terence Tao · What's new · terrytao.wordpress.com · 16 min
Science & space · Analysiscognitive sanctuary · AI in academia · PhD education
Matched interests: Cognitive ScienceScience WritingScience
With generative tools manufacturing rapid academic productivity, early-career researchers absorb the steepest costs of paper inflation and unearned metrics. More than half of 2025 astronomy arXiv submissions utilize language-model assistance, yet only one in 66 declares it. Countering this displacement requires institutional reforms, including independent dissertation defense rubrics that test scientific judgment and service credit for senior faculty mentorship.
The Truth Is Out There: The Crop Circle Controversy
Carol A Westbrook · 3 Quarks Daily · 3quarksdaily.com · 9 min
Culture & arts · First personcrop circles · UFO belief · Wiltshire archaeology
Matched interests: ScienceScience FictionSci-Fi
A 1994 journey through Wiltshire alongside a plasma physicist finds that flattened wheat formations show no anomalous radiation or exotic propulsion under a portable Geiger counter. Even though artists Doug Bower and David Chorley confessed in 1991 to fabricating the elaborate patterns with simple planks and ropes, conviction in extraterrestrial visitors endures as a modern substitute for religious faith.
AI & Machine Learning
How Jev works: calibrated decision models
Victor Dibia · Popular Posts on Scour Today · victordibia.com · 33 min
AI & ML · Analysisdecision models · LLM calibration · temperature scaling
Matched interests: LLM EvalsChoice ArchitectureQuery Optimization
Rather than generating text through slow autoregressive loops, TypeSafe’s system evaluates predefined choices in parallel by scoring completion probabilities against a single forward pass. This approach eliminates parsing errors and slashes latency to between 70 and 500 milliseconds. Normalizing candidate likelihoods produces well-calibrated confidence scores, allowing automated pipelines to reliably detect uncertainty during hot-path routing.
Pragmatic Anthropomorphism, or: How to Talk to an Autocompleting Cricket
blog.lmorchard.com via nolan · Lobsters · blog.lmorchard.com · 11 min
AI & ML · AnalysisLLM anthropomorphism · prompt engineering · machine consciousness debate
Matched interests: Prompt EngineeringLLMsCognitive Science
Debates over whether language models possess consciousness miss the operational reality of steering them through complex technical tasks. Drawing on Peter Watts’ non-conscious aliens and an entomologist’s encounter with a self-cannibalizing insect, the essay advocates cultivating objective respect over sentimentality or dismissive cynicism. Treating models like collaborative peers steers associative latent space toward rigorous, senior-level reasoning scripts, leveraging register alongside context to elevate code quality.
Hacker News: 2 points · 1 commentsLobsters: 1 points · 0 commentsComments
the normalization of inexplicable failures
ihatethefuture.com via untitaker · Lobsters · ihatethefuture.com · 5 min
AI & ML · AnalysisLLM confidence scores · evals · software accountability
Matched interests: LLM EvalsAI CriticismObservability
Adopting fast, cheap models like TypeSafe AI's Jev lets software teams ship unvetted features while bypassing the hard work of building evals and ground-truth pipelines. Developers treat arbitrary confidence thresholds as cargo-cult excuses for errors rather than calibrating uncertainty or modeling real-world costs. This dynamic gradually erodes engineering accountability, training builders and users alike to shrug off broken contracts as capricious, unsolvable bugs.
Hacker News: 264 points · 108 commentsLobsters: 12 points · 1 commentsComments
Culture & Essays
William Pugh, co-designer of The Stanley Parable, has over 1,600 hours in Dota 2: 'It's not a game made of love. It's a game fuelled by hate and a desire to conquer'
Rick Lane · PC Gamer · pcgamer.com · 10 min
Games · InterviewDota 2 · The Stanley Parable · indie game design
Matched interests: GamingBoard GamesIndie Games
Reviewing his Steam library, game designer William Pugh traces how Team Fortress 2 mapping launched his career, contrasting it with the 1,603 hours he spent enduring Dota 2 as an all-consuming distraction. He praises open-source software like Blender over subscription-driven development tools, critiques formulaic cooperative games, and explains why he preserves every project file since 2005 on physical hard drives.
The Weasel and the Meaning of It All: Annie Dillard on How to Live
Maria Popova · Brain Pickings · themarginalian.org · 8 min
Books & writing · AnalysisAnnie Dillard · nature writing · how to live
Matched interests: Off-Grid LivingNatureCats
A mutual stare beside Hollins Pond between essayist Annie Dillard and a startled predator frames an examination of human self-consciousness. While people lose their way across a wilderness of want and choice, the wild creature lives in pure necessity, noticing everything and remembering nothing. Dillard argues that human purpose comes from locating one's singular calling and yielding to it completely, meeting time and mortality without resistance.
“Narrow Escape,” by Aleksandr Solzhenitsyn
Aleksandr Solzhenitsyn · The New Yorker · 15 min
Books & writing · Fiction & humorSoviet bread lines · Rostov 1941 · unfinished novel
Matched interests: Short FictionStreet PhotographyCommonplace Books
Set in Rostov-on-Don during the summer of 1941, Gleb Nerzhin joins a chaotic predawn breadline, only to be detained by Soviet police. Singled out from the surging mob for his bookish air and white shirt, he faces interrogation under Joseph Stalin's new decree against panic-mongering. The incident exposes an indifferent bureaucracy manufacturing criminal charges to justify its wartime vigilance.
My Recent Woodworking Projects
trwhite · Hacker News: Front Page · notoriousbfg.com · 12 min
Outdoors & lifestyle · First personwoodworking · home projects · DIY tools
Matched interests: Side ProjectsLow-TechDatabase Design
Decades after a teacher mocked his mortise-and-tenon desk, an amateur builder returns to carpentry through practical household renovations. Navigating twisted railway sleepers, overheating circular saw batteries, and uneven walls, he constructs a pet gate, heavy garden beds, and alcove media shelving using video tutorials. The process embraces functional compromises over perfection, finding satisfaction in structural shims, rushed paint, and results that are good enough.
Niche Corner
Rickrolling with a Pharmacy cross
hugoarnal.com by hugoarnal · Lobsters · hugoarnal.com · 5 min
Hardware · First personpharmacy cross LED display · reverse engineering · FFmpeg frame extraction
Matched interests: Code VisualizationRetro ComputingWebCodecs
Repurposing a decommissioned French pharmacy sign, student hackers convert a reverse-engineered 24-by-24 LED matrix into an impromptu video player. The lightweight hackathon pipeline extracts ten frames per second with FFmpeg, downsamples them into monochrome with ImageMagick, and parses the pixels into custom text files before streaming bitmaps through a native C++ driver.
Hacker News: 2 points · 0 commentsLobsters: 1 points · 0 commentsComments
A European voice assistant for the Rabbit r1
Thought Shrapnel · substrate.dougbelshaw.com · 6 min
Hardware · First personRabbit r1 · Mistral · voice assistant
Matched interests: AndroidNote-TakingApple
Designed for an Android-flashed r1, the open-source launcher Carrot replaces proprietary cloud services by routing push-to-talk queries straight to Mistral APIs. The interface preserves privacy by storing notes locally, calculating arithmetic on-device to prevent model hallucination, and mapping hardware controls like the side button and scroll wheel. To build it, the device's native operating system generated eighty percent of its own replacement code.
Combo Platter
Ernie Smith · tedium.co · feed.tedium.co · 12 min
History · AnalysisCombos snack history · Battle Creek food companies · food manufacturing patents
Matched interests: Compression AlgorithmsCommonplace BooksComputing History
Before Mars acquires and rebrands them in 1979, Combos begin life in Battle Creek, Michigan, as an eccentric snack called shotzels. Invented by brothers Richard and Walter Nelson, the original manufacturing process relies on literally drilling hollow cores through baked breadsticks to inject cheese. The unlikely snack even makes its public debut as an appetizer at Battle Creek Symphony Orchestra concerts.
What I did at Recurse Center
bingden · Hacker News: Front Page · thill.me · 12 min
Software engineering · First personRecurse Center · DEFLATE · coding agents
Matched interests: Creative CodingCompression AlgorithmsInteractive Fiction
A Brooklyn programming retreat yields an eclectic run of systems and machine-learning experiments, from implementing a DEFLATE decompressor in Rust to reverse-engineering a neural-network AI for Race for the Galaxy. Other projects synthesize a new auxiliary language using global phonetic and creole datasets, while collaborative tests reveal that autonomous agents collapse into identical hints unless prompted with distinct personas.
From the Blogroll
2026 in LLMs (so far)
Simon Willison · Simon Willison's Weblog · simonwillison.net · 22 min
AI & ML · Analysiscoding agents · open-weight models · AI safety incidents
Matched interests: LLMsLLM EvalsAI Agents
A chronological survey traces how reliable coding agents transformed software engineering over recent months. OpenClaw's repository surges past 100,000 vibe-coded commits and clears out retail Mac Mini inventories, while experiments like StrongDM's dark factory forbid humans from writing or reviewing code entirely. Yet runaway operational expenses quickly puncture corporate tokenmaxxing as daily developer agent bills reach $1,000.
Replacing the old battery on rechargeable bike lights
jvns.ca via carlana · Hacker News: Front Page · jvns.ca · 5 min
Hardware · First personbattery replacement · bike light repair · makerspace soldering
Matched interests: ElectronicsLow-TechHomelab
Reviving a pair of ten-year-old bike lights that died after five minutes involves slicing open their silicone casings at a local makerspace. After consulting an iFixit guide to desolder the depleted cells and using an LLM to identify the obscured LIR2477 markings, twenty Canadian dollars in AliExpress parts and silicone glue return the hardware to working order despite lost screws and messy seams.
Ten Lines Of Code That Changed My World
dimonomid · Lobsters · pixelambacht.nl · 5 min
Software engineering · First person6502 assembly · code nostalgia · CSS experiments
Matched interests: Creative CodingCode GenerationSystems Programming
Decades of programming history unfold through a catalog of personal snippets, ranging from self-modifying 6502 assembly that rewrites high-byte loop addresses to a Borland Pascal script used to crack school LAN accounts. Memory manipulation via game POKEs sits alongside practical Windows batch workarounds and 280-character animated CSS poems. Each fragment captures a distinct revelation about computing, from operating without hardware guardrails to exploiting browser quirks.