---
title: "Nov 16: Who owns AI-generated code, and more"
show: "RobotPaper's daily HN podcast"
canonical: "https://podcasts.robotpaper.ai/hn/ep/2026-11-16/"
date: "2026-11-16T03:32:09+00:00"
duration: "29:44"
duration_seconds: 1784
audio: "https://podcasts.robotpaper.ai/hn/ep/2026-11-16/2026-11-16-hn.mp3"
chapters: "https://podcasts.robotpaper.ai/hn/ep/2026-11-16/chapters.json"
transcript: "https://podcasts.robotpaper.ai/hn/ep/2026-11-16.txt"
feed: "https://podcasts.robotpaper.ai/hn/feed.xml"
language: "en"
author: "Roy Osherove"
ai_generated_voices: true
---

# Nov 16: Who owns AI-generated code, and more

RobotPaper's daily HN podcast · Nov 16, 2026 · 30 min

- Listen: [MP3, 29 MB](https://podcasts.robotpaper.ai/hn/ep/2026-11-16/2026-11-16-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/2026-11-16/
- Chapters (JSON): https://podcasts.robotpaper.ai/hn/ep/2026-11-16/chapters.json
- Transcript (text): https://podcasts.robotpaper.ai/hn/ep/2026-11-16.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
- `00:00:33` AI code ownership
- `00:02:56` MacBook OLED flicker
- `00:04:58` Nvidia financing
- `00:07:14` DRAM supply lawsuit
- `00:09:32` Software job titles
- `00:11:47` Chinese chip access
- `00:15:47` Ambient recording
- `00:17:58` Rust library safety
- `00:20:04` Pool chlorine alerts
- `00:22:13` Disclosure deadlines
- `00:24:21` Jeep computer access
- `00:24:59` Models on old devices
- `00:25:38` FM-1 Workbench
- `00:26:14` Claude login outage
- `00:26:51` Misleading job ads
- `00:27:29` Woz writes BASIC
- `00:28:06` Gemini orchestration
- `00:28:47` Monkey Island editions
- `00:29:21` Outro

## Show notes

- [All AI-generated code should be Open Source](https://jakub.jankiewicz.org/blog/ai-code-open-source/) · [HN discussion](https://news.ycombinator.com/item?id=50507303) \| quoted: [Aurornis](https://news.ycombinator.com/user?id=Aurornis)
- [The MacBook Pro's OLED PWM problem](https://macanorak.com/the-macbook-pros-oled-pwm-problem-when-a-better-display-becomes-worse-to-use/) · [HN discussion](https://news.ycombinator.com/item?id=50506719) \| quoted: [Telaneo](https://news.ycombinator.com/user?id=Telaneo)
- [Nvidia deliberately kept GPU prices high through circular financing deals](https://www.techspot.com/news/114558-nvidia-deliberately-kept-gpu-prices-high-through-circular.html) · [HN discussion](https://news.ycombinator.com/item?id=50501631) \| quoted: [Aurornis](https://news.ycombinator.com/user?id=Aurornis)
- [Court denies motion to dismiss DRAM price fixing case](https://litnews.ai/articles/garciaguirre-samsung-[REDACTED]) · [HN discussion](https://news.ycombinator.com/item?id=50502988)
- [A family tree of software job titles](https://zwischenzugs.com/2026/11/03/a-family-tree-of-software-job-titles/) · [HN discussion](https://news.ycombinator.com/item?id=50499038)

### Chinese chip access (2 stories)

From The New York Times and Unemployment Labs, chip export rules and Blackwell fingerprints.

- [Lutnick Pushed to Weaken Chip Rules That Overlap with His Family Business](https://www.nytimes.com/2026/11/15/us/politics/lutnick-pushed-to-weaken-chip-rules-that-overlap-with-his-family-business.html) · [HN discussion](https://news.ycombinator.com/item?id=50506180) \| quoted: [sokoloff](https://news.ycombinator.com/user?id=sokoloff)
- [Chinese frontier AI models carry the fingerprint of export-controlled GPUs](https://unemploymentlabs.com/checkpoint-forensics) · [HN discussion](https://news.ycombinator.com/item?id=50474883) \| quoted: [flowerlad](https://news.ycombinator.com/user?id=flowerlad)
- [Tech companies are trying to shift the concept of privacy again](https://www.eff.org/deeplinks/2026/11/tech-companies-are-trying-shift-entire-concept-privacy-again) · [HN discussion](https://news.ycombinator.com/item?id=50496602)
- [Memory Safety Bugs in Rust Standard Library](https://forge-blog.github.io/posts/rust-std-memory-safety/) · [HN discussion](https://news.ycombinator.com/item?id=50478701) \| quoted: [Panzerschrek](https://news.ycombinator.com/user?id=Panzerschrek), [Xirdus](https://news.ycombinator.com/user?id=Xirdus)
- [Microsoft 365 outage leads to chlorine gas poisoning of 15 hotel guests](https://www1.wdr.de/nrw/rheinland/rhein-erft-kreis/technischer-defekt-chlor-phantasialand-100.html) · [HN discussion](https://news.ycombinator.com/item?id=50509411) \| quoted: [Espressosaurus](https://news.ycombinator.com/user?id=Espressosaurus), [Havoc](https://news.ycombinator.com/user?id=Havoc)
- [Project Zero will disclose vulnerabilities 90 days after reporting](https://projectzero.google/2026/11/disclosure-policy.html) · [HN discussion](https://news.ycombinator.com/item?id=50496051) \| quoted: [Wowfunhappy](https://news.ycombinator.com/user?id=Wowfunhappy)
- [Claude Code and Codex reverse engineered my 1994 Jeep's ECU over a $3 Arduino \[video\]](https://www.youtube.com/watch?v=WPj3aQ-xxpI) · [HN discussion](https://news.ycombinator.com/item?id=50463399)
- [I'm making LLMs work on small and old devices](https://github.com/hyperionaidev/smolmaxxing/) · [HN discussion](https://news.ycombinator.com/item?id=50468888)
- [FM-1 Workbench](https://open.m-vave.com/fm1-workbench) · [HN discussion](https://news.ycombinator.com/item?id=50488325)
- [Claude Is Down](https://status.claude.com/history) · [HN discussion](https://news.ycombinator.com/item?id=50481220)
- [LinkedIn, Monster take down over 1000 fake job postings after CBC investigation](https://www.cbc.ca/news/canada/linkedin-monster-take-down-over-1-000-fake-job-postings-after-cbc-investigation-9.7415379) · [HN discussion](https://news.ycombinator.com/item?id=50499723)
- [How did Woz write the Apple 1 BASIC before building the computer? (2021)](https://retrocomputing.stackexchange.com/questions/17957) · [HN discussion](https://news.ycombinator.com/item?id=50501594)
- [Gemini 4 Hydrogen and Orchestrator](https://antigravity.google/blog/gemini-4-hydrogen-in-google-antigravity/) · [HN discussion](https://news.ycombinator.com/item?id=50472260)
- [Monkey Island: all 13 versions side by side](https://monkeyisland.pages.dev/) · [HN discussion](https://news.ycombinator.com/item?id=50493904)

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 Monday, November sixteenth, 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, a proposal would make AI-generated code public. We're also covering display flicker and chip financing.

\[Grace\] 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 the Jankiewicz blog: all AI-generated code should be open source. The essay argues that developers should return generated software to the open ecosystem that supplied its training material. The author says models learned from open-source projects and attribution-licensed programming discussions, and describes turning that knowledge into proprietary software as license laundering. The post proposes explicit open-source licensing or a Creative Commons Zero public-domain dedication, particularly when human involvement consists mainly of prompting. It also discusses different approaches to computer-generated works across jurisdictions. Its proposal extends to mixed workflows in which developers accept generated code with little review or modification. The essay cites the United Kingdom as a jurisdiction that grants limited protection to computer-generated works.

\[Linus\] The comments are divided between authors seeking control over reused work and developers emphasizing their own design decisions. J R M G cites Copyright Office guidance that protects human expression within a work containing generated material, with the sufficiency of human contributions assessed case by case. That guidance also says prompts alone do not provide sufficient control under current technology. Blagie, who says they wrote multiple widely used A G P L systems, describes a company replacing improperly distributed code with a model-generated clone after receiving violation notices. Blagie says the clone preserved creative design while introducing bugs. Aurornis asks for concrete examples of models reproducing distinctive code. Decoding argues that months of direction and refinement represent substantial human authorship even when a model writes every line. Glitch C challenges the essay's licensing logic, saying a license governs rights held by a copyright owner. Tim Britt offers a practical recordkeeping approach: maintain tasks and issues documenting infrastructure, architecture, business logic and code decisions. Bilbo zero S recommends an engineering notebook, while saying that such records would not resolve a claim over generated code that reproduces somebody else's copyrighted implementation.

\[Ken\] From MacAnorak: the MacBook Pro's OLED pulse-width-modulation problem. The report examines discomfort reported by some owners of Apple's new tandem OLED display. Owners describe headaches, eye fatigue, dizziness and nausea, with some returning their machines. The article cites Notebookcheck measurements of two hundred forty hertz modulation on the fourteen-inch PROLED model at brightness settings of ninety-nine percent and below. Its comparison Mini-LED MacBook measured fourteen thousand eight hundred eighty hertz. Pulse-width modulation controls apparent brightness by varying light output rapidly. The report says modulation depth and dimming behavior also affect the display experience. It describes direct-current dimming across brightness levels on the new panel alongside the measured flicker, and reports complaints from people who previously used other OLED devices comfortably.

\[Dennis\] The thread combines reports of flicker sensitivity with disagreement over how to interpret the measurements. Legitronics examines the oscilloscope images and questions whether four hundred eighty or nine hundred sixty hertz better describes the pattern. J R M G initially agrees that the waveform spacing looks faster, then identifies the deepest troughs occurring every second dip as the basis for the two hundred forty hertz measurement. Klaus twenty-three says the deeper alternating dips create a significant lower-frequency component even when faster components are also present. Telaneo calls for more research on how many people experience these problems, describing a personal history of tolerating older displays without difficulty. Josh Triplett suggests checking room lighting by moving a hand or thin object and looking for a sequence of apparent snapshots. Bayindir H says a camera's anti-flicker warning provides another indication. M Rob describes sensitivity to the phantom-array effect during eye movements and says they choose LCD displays with flicker-free options.

\[Grace\] Next, from TechSpot: Nvidia financing deals and high GPU prices. The report describes customer financing intended to absorb semiconductor commitments and support demand for Nvidia hardware. It says the company pursued more deals in mid-to-late twenty twenty-six while facing an oversupply of ordered TSMC wafers. Colette Kress, now interim chief executive, had questioned the scale of arrangements with cloud providers and AI companies. According to the article, Nvidia later slowed, reduced or paused some deals, including an OpenAI arrangement that originally included a two hundred fifty billion dollar backstop. The report also connects data-center demand with pressure on consumer memory and graphics-card supplies. It says GeForce prices often exceeded one hundred sixty percent of their original suggested retail prices as resources shifted toward data-center GPUs.

\[Ada\] The discussion splits over whether these deals represent ordinary vendor financing or unusually concentrated financial risk. K zero morebi sees investments in prospective customers as a way to develop demand and asks what makes the practice anticompetitive. I'm down with O T P compares the arrangements with financing offered by carmakers and industrial-equipment suppliers. Part loyal demon argues that Nvidia's equity positions and guarantees expose it to correlated customers and rapidly depreciating equipment, making the structure different from separate conventional loans. Mondrian focuses on the possibility of Nvidia recording cash that originated with its own investments as customers' purchases. Hadlock connects the debate to local inference, proposing that a hypothetical two-thousand-dollar high-memory GPU could make capable models attractive at home. Aurornis counters that the computer, setup work and power costs still compete with a twenty-dollar monthly subscription, and values access from a phone. Plorkyeran adds an electricity calculation: at their rates, a four-hundred-fifty-watt card producing roughly a million tokens per hour would cost about eighteen cents per million tokens, similar to the token price they compared on OpenRouter.

\[Linus\] From LitNews: a judge denies dismissal of the DRAM price-fixing case. A proposed class action against Samsung, S K Hynix and Micron will proceed after a ruling on the sufficiency of its allegations. Judge Noël Wise found the complaint plausibly alleged coordinated restrictions on conventional memory supply as manufacturers shifted capacity toward high-bandwidth memory. The report describes long-term HBM commitments, refusals of conventional-memory contracts, delayed factories and similar customer-vetting practices among the allegations supporting that decision. The judge gave no weight to general statements about supply discipline or the concentrated market structure, and distinguished the new complaint from an earlier dismissed DRAM case. The companies' answers are due December seventh, with a case-management conference scheduled for January twelfth, twenty twenty-seven. The complaint concerns conventional memory used in computers, phones and servers, including DDR five modules.

\[Ken\] The comments mix frustration over equipment costs with debate about whether supply decisions indicate coordination. Subiculum Code says a research workstation with two hundred fifty-six gigabytes of RAM would now require four to five thousand dollars for memory, before adding a similarly expensive GPU for medical-image model training. J S C Y C suggests an institutional high-performance-computing cluster when an isolated system is unnecessary. Sigmoid ten favors rented compute for project-based training, citing access to larger machines without buying a dedicated rig. Sega Sai argues that a powerful local workstation still helps with development and experiments by avoiding cluster queues and restrictions. Subiculum Code adds that purchased lab hardware remains after a grant ends, while unused online-computing funds would be returned. G P T five questions the collusion argument when demand exceeds available inventory. Jeremy J H points to manufacturers that previously overbuilt capacity for demand that disappeared before factories opened. Klutzy Sofa stresses that the current ruling allows the case to continue and that the allegations remain to be decided by the court.

\[Dennis\] Over on the Zwischenzugs blog: a family tree of software job titles. An interactive diagram traces how technology roles have split, merged and changed names from the nineteen forties onward. It contains one hundred four titles in seven families, covering development, infrastructure, security, data, testing, support and technical leadership. Time advances in five-year intervals. Solid arrows indicate parentage, while dashed arrows indicate influence or common career moves. The author describes dates as estimates of when titles became common. The diagram is built from one Graphviz file and a JSON file of descriptions. Its examples include the division of webmaster work among several specialties, the emergence of full-stack development, and different meanings of systems engineer. Selecting a family filters the diagram, while selecting a title reveals its background and connections.

\[Grace\] Commenters recognize the recurring changes in terminology but disagree over the skills that a title should imply. Probably Wrong says a software engineer should be able to distinguish good code from bad, describing a conversation with someone publishing AI-generated Python projects. Dan Grec emphasizes requirements, design, testing, deployment, maintenance and eventual handover as parts of software engineering. A Quantized responds that implementation itself often reveals information that changes the design. Make It Double highlights the operational burden behind DevOps, including on-call rotations and incident response. N K M N Z argues for organizational structures that let development teams speak with production users, deploy changes, and learn from production behavior. Stroebs describes holding several infrastructure-related titles while repeatedly doing generalist work. There are also practical suggestions for navigating the diagram. Flurdy recommends holding Control while scrolling to zoom. Zamadatix suggests opening the PNG version when the interactive viewer's zoom controls fail, and Hello Username says that version supports pinch zoom on a phone.

\[Ada\] From The New York Times and Unemployment Labs, chip export rules and Blackwell fingerprints. Two reports examine how Chinese companies obtain access to advanced Nvidia computing. The Times describes A two oh one, a Malaysian data center scheduled to open next year with advanced Nvidia servers supplying a Chinese technology company. Its operator, Bitdeer, is a client of Cantor Fitzgerald, the financial firm formerly headed by Commerce Secretary Howard Lutnick and now run by his sons. Cantor served as a sales agent for a Bitdeer equity offering of up to one billion dollars this year. The Times reports that Lutnick has pushed for narrower restrictions on overseas data centers serving Chinese clients. Commerce's proposal would cover selected countries and future chip shipments. Companies would notify the government when Chinese customers use their facilities to train models, without automatically needing a license. The Commerce Department says Lutnick has complied with his ethics agreements. The Times reports no evidence that Cantor's investments influenced his deliberations. Unemployment Labs approaches the hardware question through published model weights. Its researchers found Blackwell initialization patterns in six models from DeepSeek, Moonshot, MiniMax and Ant Group released since April. The method recovers traces of starting weights and searches for hardware and initialization settings that reproduce them. The same researchers confirmed the matches for DeepSeek V four Flash and Kimi K three on physical Blackwell chips. The measured fingerprints identify the hardware that generated starting weights. The researchers infer that the same hardware also performed training, while the weights leave chip locations and access arrangements undetermined. Their newest readable models from Alibaba, Zhipu, Tencent, Xiaomi and StepFun still showed Hopper patterns.

\[Linus\] The comments are divided over potential conflicts of interest, export policy and the interpretation of the technical findings. In the Lutnick discussion, sokoloff argues that the business overlap is weak evidence of wrongdoing. Sokoloff points to the article's statement about the absence of evidence that Cantor's investments influenced the decisions, and says commercial policy can benefit clients of many financial firms. Brookst counters that even a small share of a portfolio can create a conflict, and argues for divestment or blind trusts to address its appearance. F B N seventy-nine questions whether restricting China's access to advanced graphics processors is a sound long-term strategy. O three march native responds that the concern is unequal treatment under the rules. In the hardware-fingerprint discussion, flowerlad emphasizes that Chinese companies can rent computing outside China, citing facilities in Singapore and Malaysia. Modeless considers the method sound and notes that the researchers' results matched the hardware disclosures companies actually made. Adrian B focuses on the difference between creating starting weights and running the training job. Adrian B describes a possible workflow in which a developer initializes a model on a Blackwell workstation, then transfers it to a cluster using other processors. Loneboat raises the same question about separating preparation from large-scale training. C M A points to the article's own explanation that weights could move between systems, or that an existing model could continue training on new hardware. Int thirty-two sixty-four says they have run large open models on cloud providers with minimal identity checks, and expects enforcement of remote access restrictions to require increasingly intrusive monitoring of customers and training workloads.

\[Ken\] From the Electronic Frontier Foundation, technology companies are trying to shift the concept of privacy again. The organization argues that wearable recording devices extend one owner's privacy choices to everyone nearby. Its examples include camera-equipped glasses from Meta and Google, Apple's Audio Intelligence for Apple Watch, and Amazon's Bee devices. These products can capture, transcribe or summarize conversations. The EFF says promises about encryption, access controls and deletion of raw audio focus on the device owner while leaving bystanders unable to opt out. Transcripts and summaries can remain after audio is discarded. The article describes these devices as continuous monitoring designed to blend into ordinary surroundings. The article also describes Google's experiments with notifying nearby phones that recording is taking place. A Google product manager said notifications can leave bystanders unsure what action to take. The EFF recommends feedback through Apple's beta feedback forms or Feedback Assistant.

\[Dennis\] The discussion combines concern about workplace surveillance with support for useful transcription and accessibility. Toddmorey says even private meetings with managers now include AI notetakers, and asks whether teams can make disabling transcription a normal request. Willis nine thirty-six describes asking teammates for objections before enabling recordings, including an invitation to respond privately. The team requested copies of the notes so people could correct them. Solar Fields cautions that removing visible meeting bots does not remove every recorder. They name Superwhisper as software that can capture audio directly from the operating system and applications, and also mention recording with a second device. Abecedarius, who describes being deaf or nearly deaf, sees substantial promise in glasses that transcribe speech. Abecedarius proposes local processing that saves or exports nothing, while noting that software could later be changed to retain those conversations.

\[Grace\] On the FORGE Blog, memory safety bugs in Rust's standard library. Microsoft researchers found flaws in unsafe library internals that can expose programs written entirely in safe Rust to undefined behavior. The team used Hyenas, its research version of the MDASH vulnerability-scanning system, alongside human investigation and reproduction. The report identifies misplaced trust in callbacks, panic behavior and the underlying system. One example involves cloning a buffer for the nightly map windows iterator. A partially constructed buffer already has an active destructor when a user-provided clone operation runs. If that operation panics, unwinding can make the destructor access uninitialized values. The fix holds the buffer in ManuallyDrop until its contents are ready. The report recommends examining every call into code outside a function's control as a possible early exit, including comparisons, destructors and iterator methods.

\[Ada\] The thread debates panic recovery and what safe interfaces must guarantee. Panzerschrek argues that immediate process termination would avoid the complexity of writing code that remains correct during unwinding. Xirdus responds that test harnesses and plugin systems may need to continue after a panic. Simonask says the final moments of a failing process can still matter for saving work and logging diagnostic information. Chai T Rex separates indexing behavior from memory corruption: ordinary indexing can panic on an invalid position, while variable dot get provides an optional result that the caller can handle. Nam J H similarly points to the slice get method for graceful handling of an index outside the valid range. Tialaramex explains that ExactSizeIterator is a safe trait whose implementations can report incorrect lengths. Code needing a trustworthy length promise can instead rely on the unsafe TrustedLen trait, whose implementer takes responsibility for that guarantee.

\[Linus\] From W D R, a Microsoft outage obscured pool chlorine alerts at Phantasialand. A dosing-system defect and undelivered warning emails preceded injuries at the Hotel Ling Bao in Brühl. Phantasialand told the broadcaster that Friday's Microsoft disruption prevented emails arriving, including alarms from the sauna area's maintenance safety system. The automatic dosing equipment had pumped too much chlorine into the pool. Fifteen guests were affected, with respiratory irritation among the reported symptoms. Four went to hospital, a correction to the initial figure of five. One remained under observation with a moderate injury. Emergency responders identified the dosing-system defect and excessive chlorine concentration. The water was replaced, and the pool reopened after its temporary closure.

\[Ken\] The comments focus on the reliability of safety alarms and responsibility across the system. N nineteen P E D L two argues that dangerous conditions should never depend entirely on email reaching a hotel employee. Dan Hulton calls for examining multiple failures together, including the system's design, staff training, regulation and inspections. Espressosaurus recommends physical alarms, with email used for convenience and management notification. Amluto says they operate an advanced chlorination system and describes how a low or zero sensor reading can cause equipment to dispense too much chlorine. Their practical recommendation is an independent measurement before the pool opens, preferably a high-quality titration drop test, with results recorded. Amluto also recommends limiting maximum dispensing rates so excessive concentrations take hours of continuous operation to develop. Havoc's translation of the German report emphasizes the two distinct failures: excessive dosing and the failed email notification. Spaldingcactus questions the thread title's reference to chlorine gas, observing that the reporting describes excessive chlorine without identifying the precise chemical exposure.

\[Dennis\] From Project Zero's blog, vulnerability disclosure is moving to fixed deadlines. Google's research team is replacing its ninety-plus-thirty-day policy with publication a set number of days after a vendor receives a report. The current deadline is ninety days. It falls to sixty days on February second, twenty twenty-seven, and thirty days on January third, twenty twenty-eight. Project Zero says AI-assisted discovery and exploitation are accelerating, while negotiating individual timelines is becoming harder as report volumes rise. The policy removes routine grace periods and the additional thirty-day patch rollout window. Exceptions remain for active exploitation, information already made public and vendor requests. The team will continue tracking fixes, but whether a bug is fixed will no longer determine its disclosure date. The article also allows exceptional handling of hardware issues, citing Spectre and Meltdown.

\[Grace\] Commenters disagree over whether tighter deadlines will improve patching or expose users of slower-moving products. Wowfunhappy accepts removing the patch rollout extension but questions imposing thirty days on bugs that are inherently difficult to fix. Sethaurus argues that deadlines balance vendors' desire for time against users' need to know about vulnerabilities, especially when attackers can discover the same flaws more quickly. Sigmoid ten highlights inexpensive Android devices, saying some vendors already need six months or more to distribute patches and may respond poorly to shorter deadlines. Peter Stuer points to bespoke on-premises systems that lack automatic updates and depend on fixes from component suppliers. Danny W says dissatisfaction with Ubuntu's vulnerability response prompted a switch to Debian. Giga chad describes running Trivy scans against Docker containers to identify which projects keep current, and says those results have influenced their choice of software exposed to the internet.

\[Ada\] A few more that didn't make the cut today. On YouTube: Claude Code and Codex reverse-engineer a nineteen ninety-four Jeep's engine computer. The creator describes connecting a three-dollar Arduino to a Cherokee without an OBD two port. Over about a week, the project logged drives, copied the computer's program and ran code inside it. The description names an Arduino Nano and ChryslerScanner parameter tables. Commenters welcomed the possibilities for repairing and controlling owned hardware, while expressing concern about agents operating physical equipment.

\[Linus\] From a GitHub project: language models on small and old devices. Smolmaxxing adapts inference to television boxes, older tablets and phones through custom kernels, memory management and weights streamed from flash. Its reported results include sixteen-billion-parameter models on an original iPad Air and a Xiaomi television box with about a gigabyte of usable memory. The project says measurements were taken on the devices themselves. Commenters admired the optimization work and reuse of older hardware, while debating the practical usefulness of models running on such constrained systems.

\[Ken\] From M-Vave: FM-one Workbench. The browser editor and accompanying manual provide controls for an eight-track instrument with six sound engines, including FM synthesis, synthesized drums and sampling. Tracks support up to sixty-four steps, and the complete HTML can be downloaded for offline playing and editing. Device transfers use Web MIDI in desktop Chrome or Edge. Commenters were enthusiastic about experimentation and custom firmware, while others questioned the hardware's connections and construction or preferred the flexibility of a laptop and MIDI controller.

\[Dennis\] From Claude's status page: a November thirteenth authentication outage. The report describes access failures affecting Claude's chatbot and coding tools, while its API continued operating. Claude Code and the developer console had partial outages, and Cowork had degraded performance. Anthropic acknowledged elevated authentication errors at nine fifty-eight U T C. Commenters expressed frustration about interrupted work and dependence on a single provider, with discussion favoring ways to continue sessions across different models and services when access fails.

\[Grace\] From C B C: LinkedIn and Monster remove misleading job advertisements. More than fifteen hundred listings were taken down after an investigation connected purported remote jobs to Evolution Travel, a multilevel-marketing company. Applicants were directed to sessions asking for upfront payments and monthly subscriptions, with income tied to travel bookings and recruitment. The advertisements omitted the company's name and enrollment fees. Commenters reacted angrily to the treatment of job seekers and described broader distrust of recruiting platforms, stale listings and online commercial incentives.

\[Ada\] From Retrocomputing Stack Exchange: how did Steve Wozniak write Apple One BASIC before building the computer? The accepted answer describes writing machine code on paper and assembling it by hand. It quotes Wozniak explaining that he could not afford a commercial time-sharing assembler, so he wrote instructions, addresses and hexadecimal values himself. A separate expression-evaluation model was tested in ALGOL. Hacker News commenters responded with memories of handwritten programs and manual assembly, while debating how early programmers moved from paper instructions to working hardware.

\[Linus\] From Google's Antigravity blog: Gemini four Hydrogen and Orchestrator. Google is introducing Hydrogen inside its coding environment alongside an option that combines it with stronger models, including Gemini four Carbon, for longer tasks. The announcement describes dynamic switching as work evolves, balancing quality, efficiency and latency. It follows Carbon's introduction with parallel agent teams. Commenters welcomed the prospect of faster routine work, while questioning whether automated model selection can judge task difficulty reliably and expressing confusion about the new model names.

\[Ken\] From the Monkey Island comparison site on Cloudflare Pages: thirteen versions side by side. The project presents the same twelve scenes across releases, using captures from ScummVM and, for Atari S T monochrome, the Hatari emulator. Images are scaled with nearest-neighbor pixels and four-to-three aspect correction to preserve their period appearance. Commenters responded with nostalgia and admiration for artwork made under tight hardware limits, while debating the visual appeal of different palettes and how much artwork changed between the original editions.

\[Dennis\] That's RobotPaper's daily HN podcast for Monday, November sixteenth. 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.
