---
title: "Nov 20: Anthropic begins Nasdaq trading, and more"
show: "RobotPaper's daily HN podcast"
canonical: "https://podcasts.robotpaper.ai/hn/ep/2026-11-20/"
date: "2026-11-20T03:34:47+00:00"
duration: "28:38"
duration_seconds: 1718
audio: "https://podcasts.robotpaper.ai/hn/ep/2026-11-20/2026-11-20-hn-v2.mp3"
chapters: "https://podcasts.robotpaper.ai/hn/ep/2026-11-20/chapters.json"
transcript: "https://podcasts.robotpaper.ai/hn/ep/2026-11-20.txt"
feed: "https://podcasts.robotpaper.ai/hn/feed.xml"
language: "en"
author: "Roy Osherove"
ai_generated_voices: true
---

# Nov 20: Anthropic begins Nasdaq trading, and more

RobotPaper's daily HN podcast · Nov 20, 2026 · 29 min

- Listen: [MP3, 28 MB](https://podcasts.robotpaper.ai/hn/ep/2026-11-20/2026-11-20-hn-v2.mp3)
- Subscribe via RSS: https://podcasts.robotpaper.ai/hn/feed.xml (paste into any podcast app)
- Web page: https://podcasts.robotpaper.ai/hn/ep/2026-11-20/
- Chapters (JSON): https://podcasts.robotpaper.ai/hn/ep/2026-11-20/chapters.json
- Transcript (text): https://podcasts.robotpaper.ai/hn/ep/2026-11-20.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
- `00:00:36` Anthropic IPO
- `00:04:11` Maintaining systems
- `00:06:28` Commission Matrix
- `00:08:21` Garage door control
- `00:10:35` Apple encryption order
- `00:12:52` Storage price outlook
- `00:14:58` Compressed Linux RAM
- `00:17:13` AI security refusals
- `00:19:12` Local AI and voice
- `00:21:08` Lastbook offline
- `00:23:11` Black-hole physics
- `00:23:47` Genie world editing
- `00:24:24` AI chip design
- `00:25:05` Ruby string overflow
- `00:25:42` Superintelligence
- `00:27:07` Making course
- `00:27:43` Email sender DNS
- `00:28:16` Outro

## Show notes

### Anthropic IPO (2 stories)

From Reuters and Anthropic: the Nasdaq debut and the offering price.

- [Anthropic’s Nasdaq debut](https://www.reuters.com/business/finance/anthropics-historic-debut-put-ais-promise-peril-wall-st-test-2026-11-19/) · [HN discussion](https://news.ycombinator.com/item?id=50562754)
- [Anthropic prices initial public offering](https://www.anthropic.com/news/anthropic-prices-initial-public-offering) · [HN discussion](https://news.ycombinator.com/item?id=50553843)
- [We need smart people who give a shit](https://violetpixel.com/blog/2026-11-17-we-need-smart-people-who-give-a-shit/) · [HN discussion](https://news.ycombinator.com/item?id=50554045)
- [How the European Commission rolled out Matrix to more than 40k staff](https://element.io/blog/how-the-european-commission-rolled-out-matrix-to-more-than-40-000-staff/) · [HN discussion](https://news.ycombinator.com/item?id=50558533)
- [Garage Door Spam](https://blog.harrison.page/garage-door-spam/) · [HN discussion](https://news.ycombinator.com/item?id=50519231) \| quoted: [datadrivenangel](https://news.ycombinator.com/user?id=datadrivenangel)
- [UK government forced to admit it issued Apple a secret encryption order](https://macanorak.com/uk-government-forced-to-admit-it-issued-apple-a-secret-encryption-order/) · [HN discussion](https://news.ycombinator.com/item?id=50562113) \| quoted: [slopinthebag](https://news.ycombinator.com/user?id=slopinthebag)
- [SSD prices may begin to decline in the latter half of 2027](https://www.drivevault.io/blog/when-will-hard-drive-prices-come-down.html) · [HN discussion](https://news.ycombinator.com/item?id=50557716)
- [Do not buy RAM, compress](https://curiousbits.nl/blog/do-not-buy-ram-compress/) · [HN discussion](https://news.ycombinator.com/item?id=50557661) \| quoted: [Lio](https://news.ycombinator.com/user?id=Lio), [jakozaur](https://news.ycombinator.com/user?id=jakozaur), [evenfrost](https://news.ycombinator.com/user?id=evenfrost)
- [Fuck Your AI Safeguards](https://ounapuu.ee/posts/2026/11/19/unsafeguards/) · [HN discussion](https://news.ycombinator.com/item?id=50562262) \| quoted: [Jach](https://news.ycombinator.com/user?id=Jach)
- [Pete warden on local AI, open models, voice interfaces, the future of hardware](https://thedataexchange.media/pete-warden-moonshine/) · [HN discussion](https://news.ycombinator.com/item?id=50561508)
- [Everything you need to survive and rebuild society](https://lastbook.org/) · [HN discussion](https://news.ycombinator.com/item?id=50560822)
- [Claude solves two open problems in the physics of black holes](https://www.anthropic.com/research/claude-black-hole-physics) · [HN discussion](https://news.ycombinator.com/item?id=50563148)
- [Genie 4 is now available](https://blog.google/innovation-and-ai/models-and-research/google-deepmind/genie-4-now-available-project-genie/) · [HN discussion](https://news.ycombinator.com/item?id=50550415)
- [Advancing chip design with OpenAI models](https://openai.com/index/advancing-chip-design-with-our-models/) · [HN discussion](https://news.ycombinator.com/item?id=50549493)
- [Ruby2ßhell: Pre-authentication RCE on stock Rails](https://verialabs.com/ruby2shell/) · [HN discussion](https://news.ycombinator.com/item?id=50556287)

### Superintelligence (2 stories)

From IEEE Spectrum and YouTube: the superintelligence and AI regulation debate.

- [The AI Superintelligence Debate Has Serious Flaws](https://spectrum.ieee.org/ai-superintelligence-debate) · [HN discussion](https://news.ycombinator.com/item?id=50559140) \| quoted: [saagarjha](https://news.ycombinator.com/user?id=saagarjha)
- [Should governments slow frontier AI? Jacob Coxon and Robin Hanson debate \[video\]](https://www.youtube.com/watch?v=lwfewYL7X0o) · [HN discussion](https://news.ycombinator.com/item?id=50564366)
- [How to make almost anything](https://pub.cba.mit.edu/863.26/site/) · [HN discussion](https://news.ycombinator.com/item?id=50514470)
- [Spamhaus new email sender requirements in November](https://www.spamhaus.org/sender-guidelines/) · [HN discussion](https://news.ycombinator.com/item?id=50523784)

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 Friday, November twentieth, 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 we're covering Anthropic's Nasdaq debut.

\[Ken\] I'm Ken, with garage-door advertising and memory compression.

\[Ada\] And I'm Ada. We're covering local models and voice interfaces.

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

\[Ken\] From Reuters and Anthropic: the Nasdaq debut and the offering price. Anthropic finished its first trading day above the offering price after retreating from its opening level. Two stories cover the listing, with Reuters reporting the trading results and Anthropic providing the terms of the offering. Reuters says shares opened at one hundred eleven dollars and fifty cents, thirty-four percent above the eighty-three-dollar offering price. They ended the day at ninety-seven dollars and seventeen cents, up seventeen percent from the offering price. Reuters distinguishes a fully diluted valuation of about two trillion dollars, including promised employee options and restricted stock, from a value based on outstanding shares that reached roughly one point five trillion dollars during trading. Anthropic's announcement lists nine hundred eighty million shares in the offering. About nine hundred fifty-six million are being sold by the company, with the rest sold by existing stockholders. Anthropic receives no proceeds from those stockholders' sales. The underwriters have a thirty-day option to buy another one hundred forty-seven million shares. The stock trades under the symbol A N T, and the offering is expected to complete today. Reuters also reports that third-quarter revenue grew forty-seven percent from the preceding quarter while the net loss narrowed. The company has nearly nine hundred billion dollars in computing commitments with partners including Amazon and Google.

\[Grace\] The two threads combine anxiety about engineering employment with disagreement over valuation and competition. In the Reuters thread, capital guy says software that can perform programming work has overturned their expectation of a secure career. Elgertam takes a more optimistic view, describing nontechnical colleagues who can begin projects with generated code but struggle once technical debt becomes substantial. Elgertam expects engineers to remain involved in delivering working systems and solving unresolved problems. P J mukh describes a team of twenty to twenty-five engineers, including six direct reports using agentic workflows. They say output has increased, but technical direction, architectural decisions and unblocking work still require accountable people. J C mont X reports the opposite staffing experience: their team went from twenty people to five over two years while shipping more than before. Two bit shifter argues that staffing decisions depend on whether a business wants more output or the same output at lower cost, and warns that managers may miss the knowledge held by the person being removed. In the offering-price thread, throwa three five six two six two compares the leading AI labs with AltaVista and Pets dot com, arguing that early market position and heavy spending do not establish a durable competitive advantage. Jump Crisscross responds with revenue multiples, putting Anthropic at forty times trailing revenue and twenty-two times annualized third-quarter revenue. Their comparison puts AltaVista at forty-three times revenue and Pets dot com at fifty-six times. Senko asks how investors could distinguish a future Google from a future AltaVista without hindsight, and points to Anthropic's ability to generate revenue from a useful product. Lord nacho focuses on competition from open-weight models. They say some colleagues prefer older models for programming and argue that increasingly capable open alternatives can constrain prices even when they remain behind the newest proprietary releases.

\[Dennis\] From the Violetpixel blog: We Need Smart People Who Give a Shit. The essay argues that fragile infrastructure depends on people who understand its failures and care enough to repair them. The author draws on work at Stripe and describes the Automated Clearing House banking system as old and burdened by exceptions, yet dependable in everyday use because people patch problems and handle unusual cases manually. The essay connects that maintenance work with two things: detailed knowledge of a system and a sense of ownership. It argues that replacing hands-on engineering with supervision of generated output can erode both. The author also links recent service outages with a loss of experienced, motivated staff, presenting that connection as their interpretation of how complex systems fail.

\[Linus\] The comments share concern about maintenance while disagreeing about the role of AI and management. George max says major infrastructure outages predate generative AI, and argues that the cost of changing software is falling faster than the cost of understanding it. They describe the time spent implementing a feature as time spent discovering dependencies and undocumented assumptions. Their practical suggestion is to use AI to explain the parts of a system that remain unclear. Marcuss two argues that many companies reward compliance over employees who challenge decisions. Chicken song replies that capable people who care about outcomes are valued on demanding projects, and says effective employees can follow direction while communicating concerns. Right byte points to the authority those teams receive, arguing that a mandate affects whether staff can act on their knowledge. Rick S addresses the pressure to operate many coding sessions at once. They distinguish running parallel tasks from reviewing their results and say serious review requires rebuilding an understanding of each problem. Floro describes a concrete source of wasted effort at their company: manual application deployments that consume hours or days, with resistance to adopting Docker or continuous integration and delivery.

\[Ada\] From the Element blog: Matrix for more than forty thousand European Commission staff. The linked report concerns a large institutional deployment of the messaging system. The rollout article could not be retrieved; the original address now returns a conference-video directory. The available reporting here comes from the fetched Hacker News discussion.

\[Ken\] The thread is divided between support for self-hosted messaging and frustration with notifications, encryption and everyday usability. C F E says their friend group runs a Matrix server with Element X clients, reporting a positive experience with notification problems reported as an exception. Razemio describes a different experience with budget Android phones, saying notifications could be absent for weeks when people did not regularly open the app. In the rollout discussion, jarek lupinski quotes the article saying that the online-and-away presence indicator was disabled as its load affected performance. Eterm questions why this feature could not scale. Jilles van gurp describes a small team's frustration with link previews being disabled by default and the difficulty of making encrypted channel history available to bots. They say the team switched to Zulip. Decentralisedog, who identifies themselves as the engineering lead for integrations at Element, details changes to previews. They say users can opt into previews for unencrypted rooms and optionally for encrypted rooms. Bundled previews let the sender supply the preview with the message so recipients do not need to fetch it themselves; the commenter says this had reached the specification and would enter the stable app shortly. Previews above the message composer let senders inspect and remove them before sending. Codedokode explains a reason to keep previews disabled: fetching a link can reveal that it was seen and can consume a one-time authentication link.

\[Grace\] From the Harrison blog: Garage Door Spam. A garage-door owner replaced advertising-filled myQ notifications with locally controlled hardware. The author says Chamberlain cut off third-party API access in late twenty twenty-three, then began sending unwanted notifications through its app. The replacement uses two ratgdo thirty-two boards, costing about sixty-two dollars each, connected to the opener's existing terminals. Their open-source firmware communicates locally through LiftMaster's Security Plus two point oh protocol and supports ESPHome or HomeKit. The author configured fixed network addresses and used the boards' JSON status endpoint to check for an open door at night, with alerts sent through Slack. The overnight check runs every ten minutes during the hours from nine in the evening through five in the morning. A Tailscale route also lets monitoring software reach each board's health-check endpoint. HomeKit supplies voice control, and the new notifications now identify who opened the door.

\[Dennis\] And the comments mix frustration with device advertising and detailed accounts of replacing vendor controls. D H C zero two says a ratgdo board connected to Home Assistant eliminated the family's need for myQ. Anamexis adds a compatibility warning: the latest Chamberlain Security Plus three point oh openers have no wired door-control connection, so that approach does not work with them. Ace Johnny two cites Apple's App Store guideline four point five point four, which requires explicit consent for promotional push notifications and a way to opt out. Ace Johnny two also identifies the Report a Problem link on an app's store page as a complaint route. Zamadatix disputes whether the notification consent described violates those terms, while datadrivenangel says promotional consent must be separate. For Android, camel Snake gives a practical control: long-press the notification, select the gear icon, then toggle individual notification categories. Beetle B recommends BuzzKill for filtering notifications.

\[Linus\] Next, from Macanorak: the UK government must acknowledge its secret encryption order to Apple. The Investigatory Powers Tribunal rejected the government's attempt to conceal the existence of a Technical Capability Notice. Macanorak reports that the original notice sought access to encrypted iCloud data globally. The government withdrew that notice after diplomatic pressure and issued another, reportedly limited to British users. Apple began a fresh legal challenge in twenty twenty-six. The tribunal found that the government had failed to establish national-security damage from acknowledging the notice in these proceedings. Its technical requirements remain secret. Liberty and Privacy International had challenged the government's secrecy position. Apple withdrew Advanced Data Protection for new UK users in February twenty twenty-five. The article says people who enabled it earlier have not been required to disable it, leaving different levels of iCloud protection depending on when the feature was activated.

\[Ada\] The thread is concerned about secrecy, with a separate dispute over what Apple's encryption statements promise. Natch argues that a change in wording leaves room for another party to possess access keys. Happy opossum challenges that interpretation, citing Apple's platform-security documentation: with Advanced Data Protection enabled, trusted devices retain sole access to the encryption keys. K P S emphasizes the practical limitation that UK users can no longer choose to enable that feature. Calibas quotes the documentation for standard protection, under which Apple stores encryption keys for recovery and only certain categories receive end-to-end encryption. On the secrecy question, slopinthebag asks how withholding these orders can be justified. Chunky nineteen ninety-four describes the government's possible argument: disclosure could prompt people under investigation to move or erase data, while also advocating later disclosure. Fweimer points to the tribunal judgment and highlights its finding that media knowledge came through confidential security briefings. Fweimer says the notice's contents remain secret.

\[Ken\] On the DriveVault blog: SSD prices may begin declining in the second half of twenty twenty-seven. The article forecasts earlier relief for flash storage than for spinning hard drives. DriveVault cites TrendForce's expectation that NAND supply will return to balance in the second half of twenty twenty-seven, following a four to five percent deficit during twenty twenty-six. The projected improvement comes from more bits produced by existing factories, additional output and weaker consumer demand. The article expects a gradual decline while overall memory prices remain elevated. Hard drives follow a different schedule because major buyers reserve output through long-term agreements that also fix prices. DriveVault says Seagate has allocated most high-capacity production into twenty twenty-eight. Its September price comparison puts a four-terabyte Samsung nine ninety Pro at six hundred nine euros, up from two hundred eighty-five euros a year earlier.

\[Grace\] And commenters disagree over whether extra manufacturing capacity or persistent AI demand will determine the next turn in prices. I mean what do I know argues that low earnings multiples for Micron and SK Hynix imply expectations of falling profits and eventually cheaper memory. Itz prime replies that cyclical manufacturers often look cheapest at peak earnings, and suggests examining price relative to tangible book value. Roofles questions whether booked factory output guarantees actual demand if planned data centers never get built. D Spillett answers that purchase contracts carry financial penalties and that buyers may keep inventory available for later projects. Rock me Amedee identifies another possible source of demand: high-bandwidth flash, using NAND beside a GPU to hold model weights. Rock me Amedee expects samples in twenty twenty-seven and possible volume production in twenty twenty-eight, and argues that adoption could put SSDs back into competition for those chips. Aurmc says hard drives retain demand because their capacity remains cheaper per gigabyte.

\[Dennis\] From the Curiousbits blog: Do Not Buy RAM, Compress. A Linux workstation experiment uses zram to fit more application data into physical memory. The author opened a large collection of Firefox tabs on a machine with thirty-two gigabytes of RAM, some reserved for graphics. The compressed swap held five point seven gigabytes of data in one point seven gigabytes of physical memory, a ratio of about three point three five to one. The article says Fedora and Pop OS enable memory compression by default, while Ubuntu does not. Windows, macOS and many Android devices already use memory compression. Compression takes processor time and can add latency. The author says compression can also extend the capacity of cloud machines. The author also describes earlier cloud applications that needed larger machines for memory despite low CPU load, and points to the Arch Linux wiki for setup guidance.

\[Linus\] The comments are enthusiastic about memory savings but divided over zram and zswap. M Z haase explains that zswap compresses pages in RAM ahead of disk swap and can move cold data to disk, while zram is a compressed block device with a fixed capacity. Mrob prefers zram with L Z four compression and wants an out-of-memory notification instead of disk swapping. Sea Brook M X reports improving a sixteen-gigabyte Framework laptop on Ubuntu by expanding the swapfile from four to sixteen gigabytes and enabling zswap with Z standard compression. Lio asks whether compression helps local model inference. Ross Bencina says dense models access every byte of their weights for each token, while mixture-of-experts models can use subsets. For browser workloads, jakozaur suggests automatically closing tabs with Tab Wrangler, and evenfrost recommends The Marvellous Suspender for suspending unused tabs. Benjiro twenty-nine suggests pressing Shift Escape in Chrome or Edge to inspect how much memory GPU operations consume.

\[Ada\] From Ounapuu: Fuck Your AI Safeguards. A software engineer describes Claude Code finding a security problem and then refusing to help repair it. The author says the issue affected their own product and needed a one-line fix. Their account describes an exemption application requiring multiple business days, identification, and the loss of data-retention exemptions. The post also describes refusals when assessing whether public proof-of-concept exploits apply to the author's technology stack. The author says EU-hosted open-weight models, including GLM five point three, eventually confirmed the vulnerability, helped fix it, and verified that the issue was resolved. The post contrasts that experience with attackers' access to models without the same restrictions.

\[Ken\] And the comments are divided between frustration with blocked legitimate work and concern about the spread of more capable unrestricted models. Skye C A says increasing refusals have prompted a shift toward open, self-hosted models, including for tasks Claude previously handled. Californical describes trying to put custom maps onto a Garmin device they own, only to encounter a refusal when asking about its sideloading restriction. Jach offers a practical observation about the interface: the same prompt appears less likely to be refused in Claude Code than in the web interface. Eddy G argues that releasing increasingly powerful open-weight models creates an irreversible proliferation problem, and says the discussion understates the reasons for safeguards. D C S S gardener compares these policies with a workplace declining to disclose an employee's next shift, because the caller's intentions are difficult to establish. D C S S gardener argues that the disputed issue is who controls the policy. W tarreau reports passing the verification program and still encountering repeated refusals when asking for help fixing bugs with an existing reproducer.

\[Grace\] From The Data Exchange, Pete Warden on local AI, open models, voice interfaces, and the future of hardware. Moonshine Voice's co-creator argues that open models are now capable enough for many tasks on ordinary laptops. In the interview with Ben Lorica, Warden points to Ollama as a straightforward way to start using them. He says capable models with around nine billion parameters are available, particularly in the Qwen family. Warden describes eight-bit weights as needing roughly nine gigabytes, with four-bit quantization bringing the model into the range of a four-gigabyte graphics card. He argues that local models face a marketing disadvantage because companies can more easily earn recurring revenue from hosted subscriptions. The interview also examines low-latency speech and specialized models.

\[Dennis\] Over in the comments, enthusiasm about control and privacy meets disagreement about the cost of dedicated hardware. L R Vick reports running eight AMD R ninety-seven hundred cards, supporting two GLM five point three Flash sessions with a million tokens of context each, at around fifty tokens per second. L R Vick says the electricity bill is about three hundred dollars a month in the Bay Area, and names Strata and Radiance as tools for running large mixture-of-experts models across consumer graphics cards. N S B K reports running Qwen three point eight Flash Next through Strata on a Mac Studio with a drag-and-drop installation. Paxy replies that cloud access can cost six dollars a month and questions the economics of buying a costly workstation. For speech interaction, N L describes a beamforming microphone array controlled by an ESP thirty-two, with Nvidia's Parakeet handling speech recognition in a browser and an L F M two point five model taking notes.

\[Linus\] From Lastbook: Everything You Need to Survive and Rebuild Society. The project packages a reference library, maps, videos, and optional local AI for use without an internet connection. It offers free, open-source applications for Mac, Windows, and Linux. Library choices range from seven point seven gigabytes for Small to sixty-seven point five for Medium and about three hundred eighty-eight for All. The optional local Gemma model adds about seventeen gigabytes and answers questions from the library with links to sources. Setup requires internet access to download the content. The listed subjects include water, food, repairs, electrical basics, and emergency preparation. Windows builds require a sixty-four-bit version of Windows ten or eleven. The Mac application requires Apple Silicon and macOS fourteen or newer, and the site says phones and tablets are unsupported.

\[Ada\] And the thread combines interest in offline information with practical criticism of the hardware requirements and the need to practice physical skills. Ferguess K says videos have been insufficient for learning manual tasks, describing a need for guided repetition with the instructor facing the same direction. Edent asks for downloadable PDFs that could run on an e-ink device with a long battery life, and argues that powering a phone would be easier than powering a computer during an emergency. My witty name suggests serving the Python application from a Raspberry Pi over Wi-Fi to several devices. Edent replies that this would require powering both the server and a phone. Pickle glitch compares the project with Project NOMAD and says they use a simple Kiwix server instead. J mathai points to Open Source Ecology for equipment designs, including houses and bulldozers. Kragen suggests cloning Lastbook's Git repository and converting the material into the desired format.

\[Ken\] A few more that didn't make the cut today. From Anthropic: Claude and two open black-hole problems. The company reports results in simplified theoretical universes used to study quantum gravity. One calculation finds an abrupt transition from hot gas to a black hole. Another establishes a bound in a modular bootstrap calculation. Anthropic says the results contribute to theoretical physics and have no practical applications. Commenters were split between interest in the mathematics and criticism of how the announcement assigns credit to the model.

\[Grace\] From Google's blog: Genie four is now available. Project Genie adds real-time editing and sound to its generated interactive worlds. Eligible Google AI Ultra subscribers can change an environment during a session by typing new prompts, including changes to weather, characters and action. Native audio adapts to the scene, including ambience and dialogue. The release also adds a community gallery. Commenters expressed enthusiasm for the technology while questioning account eligibility, commercial uses and the consequences of increasingly realistic generated worlds.

\[Dennis\] From OpenAI: advancing chip design with its models. The company describes model-assisted improvements to its Jalapeño inference chip and work with AMD on third-party hardware. Its examples include floating-point units, alternative logic implementations and changes to how calculations are divided between pipeline stages. OpenAI is also working with Synopsys to integrate models with electronic design automation tools. Commenters discussed engineering verification and the availability of the evaluated designs, while raising concerns about chipmakers sharing valuable intellectual property with a potential competitor.

\[Linus\] From Veria Labs: Ruby two Shell, pre-authentication remote code execution on stock Rails. The report describes a one-byte heap overflow in Ruby string case conversion. Rails and classic-style Sinatra expose it in their default configurations. Fixed Ruby releases are three point three point fourteen, three point four point thirteen, and four point zero point ten. Commenters focused on how a tiny memory overwrite can become a larger exploit, and debated memory safety and the difficulty of handling character conversions correctly.

\[Ada\] From IEEE Spectrum and YouTube: the superintelligence and AI regulation debate. Two stories examine expectations of advanced AI and the response to them. The Spectrum essay questions whether recursive self-improvement can overcome constraints such as obtaining useful training data outside mathematics and programming. The YouTube publisher describes an inaugural Y Combinator debate between former Anthropic and OpenAI researcher Jacob Coxon and economist Robin Hanson, on whether governments should substantially slow frontier AI. Coxon called for a slowdown, arguing that companies lack incentives to control the risks of superintelligence. Hanson favored liability insurance over broad regulation, warning about regulatory capture and expecting AI adoption to take longer. Both Hacker News threads contain sharp disagreement about risk and institutional control. In the Spectrum thread, saagarjha says the labs have responses to the objections in the article and argues that existing data may still support further improvements. In the video thread, throwaway one three three three seven argues that both companies and governments concentrate power, and favors individual ownership and control. Hadi seventy-seven I R argues for holding companies responsible for their actions and claims, including damages.

\[Ken\] From MIT: How to Make Almost Anything. The twenty twenty-six course syllabus covers a broad sequence of digital fabrication skills. Its schedule includes computer-aided design, embedded programming, three-dimensional scanning and printing, electronics production, machining, networking and project development. Supporting sessions cover version control, debugging, motor control and mesh networks. Commenters debated the value of surveying many disciplines against the depth available in each, while former students emphasized the practical projects documented on student websites.

\[Grace\] From Spamhaus: new email sender requirements in November. The organization is gradually requiring valid mail-server identification and matching forward and reverse DNS for each sending address. Every sending IP must have a reverse record pointing to a fully qualified hostname, which resolves back to that IP. The guidelines also call for confirmed opt-in for subscribers. Commenters disagreed about the burden on home mail servers and whether these measures meaningfully deter spam from large commercial providers.

\[Dennis\] That's RobotPaper's daily HN podcast for Friday, November twentieth. 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.
