---
title: "Nov 8: Apple's local family classifiers, and more"
show: "RobotPaper's daily HN podcast"
canonical: "https://podcasts.robotpaper.ai/hn/ep/2026-11-08/"
date: "2026-11-08T03:31:21+00:00"
duration: "29:35"
duration_seconds: 1775
audio: "https://podcasts.robotpaper.ai/hn/ep/2026-11-08/2026-11-08-hn.mp3"
chapters: "https://podcasts.robotpaper.ai/hn/ep/2026-11-08/chapters.json"
transcript: "https://podcasts.robotpaper.ai/hn/ep/2026-11-08.txt"
feed: "https://podcasts.robotpaper.ai/hn/feed.xml"
language: "en"
author: "Roy Osherove"
ai_generated_voices: true
---

# Nov 8: Apple's local family classifiers, and more

RobotPaper's daily HN podcast · Nov 8, 2026 · 30 min

- Listen: [MP3, 28 MB](https://podcasts.robotpaper.ai/hn/ep/2026-11-08/2026-11-08-hn.mp3)
- Subscribe via RSS: https://podcasts.robotpaper.ai/hn/feed.xml (paste into any podcast app)
- Web page: https://podcasts.robotpaper.ai/hn/ep/2026-11-08/
- Chapters (JSON): https://podcasts.robotpaper.ai/hn/ep/2026-11-08/chapters.json
- Transcript (text): https://podcasts.robotpaper.ai/hn/ep/2026-11-08.txt
- Show: [RobotPaper's daily HN podcast](https://podcasts.robotpaper.ai/hn/) (RSS: https://podcasts.robotpaper.ai/hn/feed.xml)

## Chapters

- `00:00:00` Intro
- `00:00:37` Mac knowledge graph
- `00:03:13` Asahi Linux progress
- `00:05:48` Apple keyboard filters
- `00:08:13` RAM and local AI
- `00:10:04` AI Frontiers ends
- `00:11:56` Underground in browser
- `00:13:56` Discord Rust client
- `00:16:06` Gothic migration
- `00:18:19` Libghostty stable ABI
- `00:20:22` RealPage settlements
- `00:22:24` Infected site alerts
- `00:24:21` Congress tech probes
- `00:25:01` BYD in Brazil
- `00:25:35` Hybrid model judging
- `00:26:09` Agent settings leaks
- `00:26:44` Tencent Hy4
- `00:27:18` Scientific knowledge
- `00:27:57` Ubuntu patch timing
- `00:28:32` Volunteer internet
- `00:29:10` Outro

## Show notes

- Every Mac comes with an (undocumented) family member classifier
- Asahi Linux Progress Linux 7.3
- Apple’s grey list: Words and phrases your iPhone won’t help you spell
- Upload Your RAM
- The end of Microsoft Research's AI Frontiers group
- Need for Speed Underground 2 online in the browser
- Every Windows 11 app should kill Electron, and Discord just showed why
- A barbarian tribe's mythic origin story is true, ancient DNA reveals
- Libghostty 1.0
- RealPage rent-pricing case wraps with nearly $360M to renters
- We know 15k websites are infected. how do we tell them?
- Democrats launching congressional investigations into SpaceXAI, META and Apple
- BYD doubles its Brazil market share to 9.7% a year after opening its local plant
- A Jev that can ask Opus for help gives Opus accuracy at 4x lower cost
- Claude Code's "Don't Ask Again" exposes passwords in 1,122 public GitHub files
- Tencent Hy4
- Verification Is Not Knowledge: Science needs more than answers
- Ask HN: What in the world is taking Ubuntu so long?
- From Linux to Wikipedia: How Volunteers Build the Internet

The voices you're hearing are AI-generated; the stories come from public sources linked in the show notes.\
Music: Eleven Music.\
An independent project; not affiliated with or endorsed by Amazon Web Services, Y Combinator, Hacker News or any vendor or news outlet mentioned. Opinions are the hosts' own.\
Produced by Roy Osherove.

## Transcript

\[Linus\] Good morning, it's Sunday, November eighth, and this is RobotPaper's daily HN podcast. I'm Linus. The voices you're hearing are AI-generated; the stories come from news dot Y Combinator dot com.

\[Dennis\] I'm Dennis, and today, researchers examine Apple's local family classifiers; Asahi expands Mac support.

\[Ken\] I'm Ken, with Discord's Rust client.

\[Ada\] And I'm Ada, with libghostty's stable release.

\[Grace\] I'm Grace, and as always, this podcast is generated by robotpaper dot A I, by Roy Osherove. Let's go.

\[Ada\] First up, from Eighth Rank: every Mac comes with an undocumented family member classifier. Researchers examined Apple's local personal knowledge graph and released a tool for inspecting its relationships and source records. Macs running macOS thirteen onward maintain this system, with the detailed investigation conducted on macOS twenty-six. Its SQLite databases sit behind Full Disk Access permissions. The graph combines Contacts, Calendar, Photos, Family Sharing and other sources. Separate communication logs record who communicated, when, and through which app, without message content. Two family classifiers use a hundred and forty-four features, including shared addresses, communication patterns and appearances together in photographs. Family scores are stored separately from the main graph, in a table with seventeen relationship columns recomputed daily. Retracted records can remain in an expired table, and the researchers could not establish a consistent retention period. Their iKnow dashboard uses Python's standard library, opens the databases read-only, and serves them over the local loopback connection. The authors say it can operate without an internet connection.

\[Ken\] The discussion is divided between support for useful local personalization and concern about analysis running without an explicit choice. Elgertam welcomed features that recognize faces and relationships, but wanted opt-in controls and guarantees about storage, including encryption at rest. Thomas Countz focused on social expectations: persistent records of conversations could make ordinary forgetting harder to accept and strip away the context that gives an informal exchange its meaning. Aether Spawn described the practical benefit as Siri distinguishing between contacts with the same first name, or identifying a boss when composing a message. Cyclopeanutopia reported a different concern after inspecting a kitchen MacBook with no local photos or messaging apps. While signed into an Apple account, that machine displayed contacts from iPhone messaging apps, including Signal, and image-based people clusters. Ungh reported finding the graph despite having disabled Siri and Apple Intelligence. Jungturk said other applications also use these stores, including Photos face identification and Messages deciding which messages can pass through Do Not Disturb. Jamie N K reported that a direct Settings link restored a missing Full Disk Access control on macOS twenty-seven point two.

\[Linus\] From the Asahi Linux blog: its Linux seven point three progress report. M three support is generally available, while new Thunderbolt support and a licensing change accompany the hardware work. Installation no longer requires the expert flag. Display controller support enables sleep on M three devices and use of the area beside the MacBook notch. M three graphics work has reached drawing a triangle. The shipping Thunderbolt implementation supports USB three, cross-domain connections and networking; PCI Express and DisplayPort tunneling remain unsupported. Experimental Google per-process page-size patches let the team demonstrate Windows games outside its micro virtual machine, but Asahi is withholding those patches while upstream plans and userspace integration remain unresolved. The report puts the virtual-machine performance penalty at about twenty percent, depending on the workload. Its Triforce microphone processor has also become almost twice as fast. Fedora Asahi Remix forty-five shipped alongside Fedora forty-five on November third. The project is also beginning a transition of its repositories to GPL version two or later, citing commercial forks and reuse of development code without attribution.

\[Dennis\] The thread combines enthusiasm for hardware progress with a sharp dispute over licensing and AI-assisted forks. Alsetmusic said the synchronized Fedora release and copyleft change prompted further donations. V sixty-four questioned choosing a permissive license and then objecting to unwanted uses. Palata replied that attribution was already a requirement and said the project was asking for that requirement to be honored. The discussion also contains specific installation experiences. Alentred said external USB-C displays remained the obstacle to installing Asahi on a MacBook Air. Nux reported getting a Dell monitor, including its USB and Ethernet hub, working on an M one by installing the fairydust kernel and rebooting. Nux still could not use a second external monitor. On encryption, V L Z described personally using an unencrypted boot partition with an encrypted root filesystem and linked instructions for that setup. Superdisk offered an experimental M three GPU driver that boots KDE and runs Quake and GLXGears, while saying it is not yet usable.

\[Grace\] From The Guardian: Apple's grey list of words and phrases its iPhone keyboard will not help spell. Researcher Derik Kauffman found more than a hundred thousand restricted entries in a table inside Apple's operating system. They affect predictive suggestions and swipe typing. Examples include United Ireland, Tibetan independence and the name Trayvon Martin. The list also includes the Hebrew phrase the occupied territories and the Arabic phrase the Hamas movement. The Guardian reports that swiping United Ireland can produce United Iceland, while Tibetan independence can become Tibetan independences. Restricted phrases also include iMessage is not working, alongside drug references and sexual language. Apple says its keyboard is designed to avoid inappropriate or offensive suggestions, and users can type specific phrases manually or override predictions.

\[Ada\] The comments are sharply split over whether these filters protect users from unwanted suggestions or interfere with deliberate input. Arnt said a customer once requested similar software that would avoid producing swear words on its own. Arnt argued that probabilistic swipe interpretation creates a risk of the software generating language the user never intended. Tavavex rejected that distinction, pointing out that speech recognition and touch keyboards also make probabilistic choices, and argued that the output should follow the most likely input. User seven eight three two supplied a concrete example: Ireland and United worked separately, but the same swipes together became United Iceland. P J eleven fifteen described the keyboard changing wing to Wong after swiping right wing propaganda. Repeatedly deleting and reswiping failed; the eventual repair involved double-tapping the incorrect word, deleting it, typing the replacement, tapping back to the sentence ending and pressing space. Umvi reported successfully swiping United Ireland on a Pixel. Globular Toast said FUTO Keyboard also withheld predictions for profanity. O J Ford suggested disabling autocorrection and prediction. Noah Buddy proposed training the swipe predictor solely on the user's own messages, allowing its suggestions to follow that individual's language.

\[Ken\] On the Upload Your RAM website, a satirical response to the debate over consumer memory and AI infrastructure. The page presents a fictional service for donating unused computer memory. Its premise responds to remarks attributed to Sam Altman about the utilization of memory in personal computers, and frames idle RAM as a resource consumers could donate.

\[Linus\] The discussion moves from the site's premise into a substantive argument about centralized computing, software efficiency and local control. Thih nine compared the direction of AI to the mainframe era and hoped for cheap, locally controllable frontier models. Seven nine five two saw the same pattern in cloud computing and software subscriptions: powerful local machines sitting underused while customers pay for smaller remote resources constrained by bandwidth. Peter van Dijck argued that the Altman quotation made more sense in context. Matt Rix interpreted it as a comparison of utilization rates between data-center GPUs and local machines. Carlos Jobim questioned why routine spreadsheets, documents, email and browsing now require sixteen gigabytes or more, while acknowledging higher-resolution displays and specialized workloads. Make It Double answered with concrete uses for additional memory: keeping forty browser tabs open, manipulating spreadsheets with tens of thousands of lines, and playing high-definition video while continuing other work. D O F M proposed an offline comparison using Gemma four twelve B Q A T on a machine capable of running it: disconnect the internet and examine what the computer can still do. D O F M added that the smaller E four B model could demonstrate the change too.

\[Dennis\] On Hunch, the end of Microsoft Research's AI Frontiers group. A founding researcher describes leaving Microsoft after a reorganization shifted the team's work toward model scaling. AI Frontiers formed in twenty twenty-three and had about fifty people. Its projects included the AutoGen orchestration framework, the Fara computer-use agent, and research on recursive transformers and optimization. The author says the group moved into Microsoft AI in early September, where leadership prioritized scaling and data work. He says that direction conflicted with his research interests, and that other founding leaders reached similar conclusions about leaving. The group had also shared a monorepo across Microsoft Research, including a model pretraining pipeline. The post says internships with the former group will not be available in the coming year.

\[Grace\] The thread is divided over continued scaling and support for exploratory research. Another Good Name argues that companies should test models with ten trillion and then a hundred trillion parameters while the existing approach keeps improving. W tetzner responds that exploring other architectures can still produce models with less demanding hardware requirements. Augment me says data is becoming scarcer than compute, leading teams to train on more tokens than the optimal balance would suggest. Sheepscreek identifies research directions that could coexist with scaling: latent reasoning, and separating reasoning ability from stored factual knowledge so models can retrieve specialized information through tools. A C Count thirty-nine argues that proposed replacements have so far produced incremental transformer improvements or architectures that are difficult to use without better performance.

\[Ada\] From the Underground Two Revived site, Need for Speed Underground Two online in the browser. A fan project recompiles the PC game's code for Chrome and Firefox and adds online racing. Its documentation lists widescreen play at up to a hundred and twenty frames per second, multiplayer free roam, and support for controllers, wheels and touch controls. The developer says the project began as a way to play again with a brother. Its code derives from the October sixteenth MIT-licensed version of N F S U Two Recomp, an x eighty-six to C static recompiler. The site supports cloud-save synchronization between devices and downloadable save backups. Saving during play uses Escape or Start, then Garage, then Save. The source code is published on GitHub. The site identifies itself as a fan project unaffiliated with Electronic Arts. The project accepts donations, with the same game features available to everyone.

\[Ken\] The comments combine enthusiasm for the port with concern about preservation and redistribution rights. Mac N C H Z reports that the game worked on a phone after years of remembering it on a friend's gaming PC. M randish proposes expanded fair use after twenty years to permit noncommercial sharing of old game code while treating franchise characters separately. Basch asks how such a system would handle separately licensed music, engines and characters. Mayeul C points to a directory of game recompilations, while noting its missing search and contribution features. Leni five thirty-six reports good recent results with Lutris installation scripts for the original game on Linux, following earlier versions that frequently crashed. Birdsongs provides a specific way into the multiplayer environment: start career mode, where other players appear and races and lobbies are available. Birdsongs also recommends enabling ghost mode when many other cars are present.

\[Linus\] From Windows Latest, Discord demonstrates its Rust desktop rebuild. An early client called Discord Phoenix loads faster in a cold-start demonstration and supports concurrent accounts. Discord's chief technology officer says a Task Manager snapshot showing two hundred ninety point two megabytes included two signed-in accounts. The desktop interface uses G P U I, while YouTube embeds and Activities retain system webviews. Discord says updates run in the background, removing the startup update modal. Mobile interfaces will use native components with shared Rust logic, and the browser version will retain HTML rendering with WebAssembly logic. Private testing is planned, and Discord has not announced a public release date. The report also says the new client stores login credentials in the system keyring, including Windows Credential Manager.

\[Dennis\] The thread balances performance enthusiasm with concerns about development cost and accessibility. John H H N, who describes being legally blind, says per-application interface zoom has made software much easier to use. Doohickey D highlights Electron's existing accessibility and support for non-English input methods, and questions whether Rust toolkits provide comparable behavior. Jenadine says G P U I follows Zed's requirements, with breaking changes and declined outside contributions leading to separate forks. A favour recommends Mozilla's Uni F F I for connecting shared Rust libraries to native clients. Caspper sixty-nine describes another working arrangement: a Rust backend connected to a C sharp Avalonia interface using Protocol Buffers, named pipes on Windows, and domain sockets on Linux and macOS. Caspper sixty-nine says C sharp ahead-of-time compilation produces self-contained applications, and reports coverage across Windows, Linux and macOS on both x sixty-four and ARM sixty-four, with Avalonia providing accessibility support.

\[Grace\] From Science, ancient DNA supports the Goths' Scandinavian origin story. Genetic studies trace substantial migration into northern Poland, followed by mixing and cultural transmission farther south. In five fifty-one C E, the historian Jordanes described his people's ancestors sailing from Scandinavia and later moving toward the Black Sea. Science reports that a Nature study found a major shift toward Iron Age Swedish ancestry in northern Poland around the appearance of Gothic artifacts, about fifty C E. In southeastern Polish burials dating from one seventy-five to three fifty C E, roughly half the people sequenced had Scandinavian-like ancestry. A father and daughter shared long stretches of DNA, while the daughter's maternal ancestry came from another population. A separate study found Northern European ancestry in about a third of Gothic and Germanic-style burials across Spain. Researchers describe an initial large migration followed by interbreeding and the spread of Gothic traditions through wider confederacies.

\[Ada\] The comments debate how historians weigh migration, cultural exchange and ancient accounts. Joe Gibbs asked why a Scandinavian origin had seemed doubtful, pointing to Gotland, Gothenburg and the Geats. Senderista connected earlier skepticism to the postwar rejection of racist archaeology and argued that each case needs evidence distinguishing migration from cultural diffusion. Senderista used Bell Beaker culture as an example, describing cultural transmission eastward from Iberia followed by migration back toward the Atlantic. Bazoom forty-two emphasized the article's geographical distinction: substantial migration into northern Poland, then smaller elite movements and cultural transmission farther south. In a separate exchange, C M A two fifty-six attributed Gothic movement to pressure from the Huns. T F N A replied that the Huns arrived centuries after Gothic expansion from Scandinavia and distinguished that earlier expansion south and east from the later movement west into the Roman Empire.

\[Ken\] On Mitchell Hashimoto's blog, Libghostty one point oh. Ghostty's terminal-emulation library now has a stable C interface and binary compatibility policy. Hashimoto says JetBrains terminals already use it, while Neovim's approved replacement for libvterm was awaiting a tagged release before merging. That change addresses scrollback reflow when a terminal is resized. The library uses opaque handles and extensible getters and setters, with compatibility checks enforced in continuous integration. Signed source and dependency archives support completely offline distribution builds. Its default build links libc, and disabling SIMD permits a freestanding build. The library creates no threads and supports a customizable allocator. Hashimoto reports running ten thousand active terminals on a four-core Linux server with eight gigabytes of RAM, using dash because bash exceeded the memory budget.

\[Linus\] The thread is enthusiastic about the library, with detailed comparisons of terminal features and C interface design. S M J Edison, building an interpreter in Zig, asked for examples of C APIs that hide implementation details. Spacechild one recommended libopus and the A R A API, explaining that a structure can put its size first so the receiver can check whether later members exist. Agosz suggested version and size as the first fields, with new fields appended at the end. Agosz said callers can request the structure size, allocate a buffer and use the fields that they need. Paradox four sixty explained a practical iTerm two feature: starting or resuming tmux with dash capital C capital C integrates tmux tabs and splits into iTerm's graphical interface. Amterp described building Muster because existing tools such as herdr and cmux did not match the agent workflow they wanted. Libghostty supplied the terminal foundation while Amterp concentrated on agent support.

\[Dennis\] From Multifamily Dive, the RealPage rent-pricing case reaches nearly three hundred sixty million dollars in settlements. A federal court approved agreements with landlords accused of coordinating rents through shared pricing software. The settlements cover multifamily renters at properties using named RealPage products, including YieldStar and Lease Rent Options, during the specified period from October eighteen, twenty eighteen, through November twenty-one, twenty twenty-five. The companies agreed to stop supplying nonpublic data for pricing recommendations and to stop using competitors' nonpublic data in those recommendations. Several state attorneys general opposed the agreements over their separate lawsuits, and Judge Waverly Crenshaw rejected their request to block approval. The dismissal preserves claims against defendants who did not settle. Similar state and local lawsuits remain ongoing.

\[Grace\] The comments are divided between criticism of coordinated pricing and arguments that housing supply drives rents. Rabid Heuristics argued that RealPage's accumulated historical pricing data could continue supporting coordination after the settlement. Consensus one called the case weak and characterized the software as price discovery in an undersupplied market. Fire Beyond challenged that position with quoted training material about limiting overrides of recommended rents. Fire Beyond cited an internal Greystar presentation seeking acceptance of at least ninety-five percent of RealPage-generated prices, alongside requirements for a written business justification when proposing an override. Epistasis proposed a public rental-price registry to reduce information disparities, while saying it would address individual overcharging without systematically lowering prices. George L said they had filed a claim and that the settlement calculator estimated roughly fifteen hundred dollars across three named landlords, depending on receiving the full amount.

\[Ada\] On Scott Helme's blog, how to notify thousands of infected websites. Report URI says it has identified fifteen thousand five hundred seventy-nine sites running malicious JavaScript, with an estimated combined audience exceeding one hundred thirty-five million monthly visits. The infections include payment-card skimmers, ClickFix malware prompts and malicious redirects. Entry routes include compromised WordPress installations, third-party scripts, cross-site scripting and expired script domains bought by attackers. Helme says direct notification gets few responses, security dot text adoption is low, and content delivery networks obscure hosting providers. His update says a public list could expose vulnerable sites to more attackers and leave repaired businesses struggling to escape blocklists. He is offering regional infection lists to national cybersecurity agencies and incident-response teams.

\[Ken\] The comments split over immediate browser warnings and coordinated notification of the affected website owners. Warkdarrior proposed reporting sites to Google Safe Browsing. Aachen, who runs a file-upload website, described the difficulty of removing a domain from numerous blocklists after harmful content appears. M X Uribe suggested identifying hosting providers and consolidating reports for companies such as O V H, Hetzner, DigitalOcean and HostGator, allowing providers to contact their customers. Two Thousand Ultra Deluxe said the hosting provider's abuse mailbox works, while warning that clearing ClickFix infections can require taking a site offline. Scott Helme said site owners could use the contact email in his post to ask whether they were listed. M C eachen proposed a dot well-known or DNS handshake, similar to certbot, to prove domain ownership before releasing an infection report.

\[Linus\] A few more that didn't make the cut today. From The Washington Post, congressional investigations into major technology companies. Democrats are preparing inquiries into Apple, Meta and Elon Musk's companies after winning House and Senate majorities. The reported targets include Grok's use in planning the Maduro operation, tracking apps used by immigration enforcement, and data center nondisclosure agreements. The comments mix support for corporate accountability with doubts about lawmakers' technical preparation and concern that hearings could become partisan performances.

\[Dennis\] From Bloomberg, BYD doubles its Brazilian market share. The HN headline reports a share of nine point seven percent a year after the company's local plant opened. The full article body was unavailable. The discussion is divided between enthusiasm about affordable Chinese electric vehicles and concern about subsidies, labor conditions and vehicle data collection. Commenters also debate the effect of tariffs on competition and whether lower purchase prices translate into lower ownership costs over time.

\[Grace\] From the Hightouch blog, combining Jev and Opus for identity checks. The company tested whether cheaper models could judge incorrectly merged customer records and refer uncertain cases to Opus. On its easier synthetic dataset, the hybrid achieved ninety-eight point eight percent accuracy at eighty-eight cents per thousand identities, compared with three dollars eighty-six for Opus. Savings were smaller on the harder test. The comments combine interest in model routing with concern about profiling and whether the results transfer to other tasks.

\[Ada\] From the Apiiro blog, Claude Code permission settings expose secrets when published. Apiiro found credentials in public GitHub repositories containing saved agent commands and settings. Choosing don't ask again can preserve a command containing a password in a local settings file, which developers then commit. The researchers did not test the credentials; an affected organization confirmed a key had been active. The discussion is divided over responsibility for secret handling, repository ignore rules and the security of autonomous agent permissions.

\[Ken\] From Tencent, Hy four adds vision and computer use. Tencent says the model can interpret images and charts and operate across graphical and command-line interfaces. Training includes workflows involving Blender and CAD, plus game development work with Tencent's studios. The model is available through Tencent Cloud and OpenRouter, with an open-weight release planned. The comments show interest in its visual demonstrations and game development capabilities, alongside debate about competitive pricing, surveillance applications and the gap with proprietary models.

\[Linus\] On the Cosmos Institute blog, Verification Is Not Knowledge. Jessica Hullman argues that scientific progress requires people who can challenge explanations, examine assumptions and take responsibility for assertions. Her essay considers whether AI systems could become accountable through persistent identities and durable corrections, while arguing that humans must retain those critical skills. The comments divide between valuing verified answers regardless of how they were found and insisting that understanding supports better questions, explanations and further scientific work.

\[Dennis\] Posted on Hacker News, a question about Ubuntu's security update delays. Lin somniac describes waiting nearly a week for a kernel fix after applying mitigations, and asks whether kernel variants, testing or staffing explain the wait. The post describes its security tracker entry as work in progress at the time of writing. The comments debate the balance between urgent fixes and kernel validation, with broader disagreement over Canonical's management, confidence in long-term support releases and alternative server distributions.

\[Grace\] From the MIT Press Reader, how volunteers build the internet. An essay adapted from Peer Production examines Linux, Wikipedia and other projects organized around shared contributions and public resources. It describes varied motivations, participant governance and decentralized work, while also discussing commercial participation, emerging hierarchies and barriers facing newcomers. The comments combine appreciation for volunteer work with arguments about paid maintenance, corporate influence and whether today's platforms leave people with the time, skills and independence to contribute.

\[Dennis\] That's RobotPaper's daily HN podcast for Sunday, November eighth. This podcast is generated by robotpaper dot A I, by Roy Osherove. For more podcasts from RobotPaper, head over to podcasts dot robotpaper dot A I.

---

The voices you're hearing are AI-generated; the stories come from public sources linked in the show notes. An independent project; not affiliated with or endorsed by Amazon Web Services, Y Combinator, Hacker News or any vendor or news outlet mentioned. Opinions are the hosts' own. Produced by Roy Osherove. Privacy: We count visits to these pages with Cloudflare Web Analytics (cookieless). Google Analytics (which sets cookies): visitors in the EU/EEA, the UK and Switzerland are asked first and it runs only if they accept (we go by the country Cloudflare reports for the connection; if it is unknown, we ask). Elsewhere it loads by default. Anyone can turn it off with the Privacy settings button at the bottom of every page; the choice is kept for a year in this browser (local storage "analytics-consent"). We count feed fetches and downloads in aggregate; IP addresses are hashed and never stored. If you send feedback, we keep what you write (and your email if you give it) to read and reply. © 2026 Roy Osherove.
