---
title: "Nov 22: Lean verifies the Riemann proof, and more"
show: "RobotPaper's daily HN podcast"
canonical: "https://podcasts.robotpaper.ai/hn/ep/2026-11-22/"
date: "2026-11-22T03:33:09+00:00"
duration: "29:55"
duration_seconds: 1795
audio: "https://podcasts.robotpaper.ai/hn/ep/2026-11-22/2026-11-22-hn.mp3"
chapters: "https://podcasts.robotpaper.ai/hn/ep/2026-11-22/chapters.json"
transcript: "https://podcasts.robotpaper.ai/hn/ep/2026-11-22.txt"
feed: "https://podcasts.robotpaper.ai/hn/feed.xml"
language: "en"
author: "Roy Osherove"
ai_generated_voices: true
---

# Nov 22: Lean verifies the Riemann proof, and more

RobotPaper's daily HN podcast · Nov 22, 2026 · 30 min

- Listen: [MP3, 29 MB](https://podcasts.robotpaper.ai/hn/ep/2026-11-22/2026-11-22-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/2026-11-22/
- Chapters (JSON): https://podcasts.robotpaper.ai/hn/ep/2026-11-22/chapters.json
- Transcript (text): https://podcasts.robotpaper.ai/hn/ep/2026-11-22.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
- `00:00:32` Riemann verification
- `00:06:36` Engineering job search
- `00:08:42` Iowa explosives delay
- `00:10:50` Nvidia investigations
- `00:13:51` Inkling 2 Mini
- `00:16:07` Monitor Ethernet loss
- `00:18:24` Jujutsu workflows
- `00:20:46` Coding and review
- `00:22:57` Immigration questions
- `00:25:13` Pancreatic cancer pill
- `00:25:50` Simple Bench results
- `00:26:27` Cancer virus trial
- `00:27:03` Home computing debate
- `00:27:39` Synology OAuth flaw
- `00:28:19` UK fridge security
- `00:28:57` PHP language update
- `00:29:31` Outro

## Show notes

### Riemann verification (3 stories)

From the Leanprover Community blog on GitHub Pages: Lean verification of the Riemann Hypothesis.

- [Lean verification of the Riemann Hypothesis](https://leanprover-community.github.io/blog/posts/riemann-hypothesis/) · [HN discussion](https://news.ycombinator.com/item?id=50589618)
- [Unforced Navier-Stokes Singularity and the Generalized Riemann Hypothesis](https://terrytao.wordpress.com/2026/11/20/unforced-euler-singularity-unforced-navier-stokes-singularity-and-the-generalized-riemann-hypothesis/) · [HN discussion](https://news.ycombinator.com/item?id=50578358)
- [Re: Riemann](https://borretti.me/article/re-riemann) · [HN discussion](https://news.ycombinator.com/item?id=50583982) \| quoted: [Kuyawa](https://news.ycombinator.com/user?id=Kuyawa)
- [Ask HN: I have been unemployed for 2y. 15 years of experience, US. What to do?](https://news.ycombinator.com/item?id=50583761) · [HN discussion](https://news.ycombinator.com/item?id=50583761)
- [Detonating a pile of lead azide](https://www.science.org/content/blog-post/detonating-huge-pile-lead-azide-really) · [HN discussion](https://news.ycombinator.com/item?id=50588982) \| quoted: [Loughla](https://news.ycombinator.com/user?id=Loughla)

### Nvidia investigations (2 stories)

From The Economist, Jensen Huang's fall is bad news for Silicon Valley's accelerationists.

- [Jensen Huang's fall is bad news for Silicon Valley's accelerationists](https://www.economist.com/business/2026/11/19/jensen-huangs-fall-is-bad-news-for-silicon-valleys-accelerationists) · [HN discussion](https://news.ycombinator.com/item?id=50591835)
- [US SEC's Nvidia probe is examining potential accounting issues – sources](https://www.marketscreener.com/news/us-sec-s-nvidia-probe-is-examining-potential-accounting-issues-sources-say-ce785ed3dd8ff627) · [HN discussion](https://news.ycombinator.com/item?id=50581438)
- [Inkling 2 Mini](https://thinkingmachines.ai/news/inkling-2-mini/) · [HN discussion](https://news.ycombinator.com/item?id=50581260)
- [My monitor was cutting my download speed by 75%](https://blog.tymscar.com/posts/studiodisplayethernet/) · [HN discussion](https://news.ycombinator.com/item?id=50587875) \| quoted: [Firehawke](https://news.ycombinator.com/user?id=Firehawke)
- [Jujutsu Primer](https://ethanlowman.com/posts/jj-primer/) · [HN discussion](https://news.ycombinator.com/item?id=50570628)
- [If you stop writing code, you also get worse at reviewing it](https://cekrem.github.io/posts/if-you-stop-writing-code-you-also-get-worse-at-reviewing-it/) · [HN discussion](https://news.ycombinator.com/item?id=50573886)
- [I'm Peter Roberts, immigration attorney who does work for YC and startups. AMA](https://news.ycombinator.com/item?id=50577613) · [HN discussion](https://news.ycombinator.com/item?id=50577613)
- [Pill to treat pancreatic cancer hailed as 'biggest breakthrough in decades'](https://www.theguardian.com/society/2026/nov/20/pill-pancreatic-cancer-breakthrough-daraxonrasib) · [HN discussion](https://news.ycombinator.com/item?id=50582069)
- [Astra 6.2 and Sol 6.2 flop at Simple Bench](https://simple-bench.com/) · [HN discussion](https://news.ycombinator.com/item?id=50588916)
- [Pancreatic cancer halted by virus injection in three patients](https://www.newscientist.com/article/2528235-pancreatic-cancer-halted-by-virus-injection-in-three-patients/) · [HN discussion](https://news.ycombinator.com/item?id=50563066)
- [Software companies are trying to kill home computing?](https://lateblt.livejournal.com/526188.html) · [HN discussion](https://news.ycombinator.com/item?id=50554010)
- [Synology Security Advisory](https://www.synology.com/en-uk/security/advisory/Synology_SA_26_21) · [HN discussion](https://news.ycombinator.com/item?id=50567209)
- [UK fridges secured from hackers](https://www.bbc.co.uk/news/articles/c6eq49zyx7p4o) · [HN discussion](https://news.ycombinator.com/item?id=50572430)
- [What's New in PHP 8.6](https://www.php.net/releases/8.6/en.php) · [HN discussion](https://news.ycombinator.com/item?id=50557870)

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 twenty-second, 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, the Lean community verifies a Riemann proof.

\[Ada\] And I'm Ada. We also have local AI and network troubleshooting.

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

\[Ada\] From the Leanprover Community blog on GitHub Pages: Lean verification of the Riemann Hypothesis. The Lean and Mathlib communities say independent checks have confirmed both the mathematical statement and Levent Alpöge's proof. Three stories today cover that result, related mathematical breakthroughs and the response to AI-assisted discovery.

The joint statement says Alpöge released a Lean library containing seventeen million lines of code on November twentieth. Its direct dependency is Mathlib. Experts in Lean and number theory had already vetted Mathlib's statement of the Riemann Hypothesis. Dozens of people have now reported successful compilation, reproducing the same raw proof export.

The community also isolated the statement in a file importing only Mathlib. Two independent tools, comparator and con-leche with its compare option, confirmed that the theorem in Alpöge's library matches that statement exactly. The exported proof passed Lean's official kernel, con-leche, con-ron and other checkers implemented in different programming languages. The statement says con-leche and con-ron have formal correctness proofs relative to set theory. These checks operated on an inert export, beyond the reach of executable code in the proof sources.

On Terence Tao's WordPress blog, the accompanying report covers unforced Navier-Stokes singularity and the Generalized Riemann Hypothesis. Tao describes different stages of mathematical exposition. Alpöge, Tristan Buckmaster and Matei Coiculescu have produced a human-readable, Lean-formalized proof of finite-time blowup for the three-dimensional unforced incompressible Euler equations, with heavy AI assistance.

Alpöge and Buckmaster also produced a Lean-formalized result for the corresponding Navier-Stokes equations. Tao describes its strategy as constructing an approximate discretely self-similar solution numerically, then applying rigorous perturbation theory to obtain an exact solution. He says the paper needs further work to become a professionally written proof. The Generalized Riemann Hypothesis result uses an internal AI model and has an enormous Lean-only proof whose strategy researchers are still extracting.

Borretti's blog responds in an essay titled Re: Riemann. It quotes Alpöge describing how he guided the combination of two ideas from a toy setting before Claude outpaced him near the finish. Borretti objects to a future that places human beings at the margins of scientific discovery. The essay contrasts that prospect with accounts of human agency in the Apollo eleven launch and Diego Rivera's mural, Man, Controller of the Universe.

\[Dennis\] The discussions combine confidence in the verification with disagreement about understanding, reuse and the human role in discovery. In the Lean community thread, R C R anti argues that independently implemented checkers make a shared exploitable bug increasingly unlikely. Sunshowers explains that formal verification still depends on a trusted computing base. Smaller bases and multiple independent implementations can increase assurance.

The user the joke is on me raises an earlier fake Collatz refutation that passed independent checkers and questions the community's confidence. Two F X responds that Lean now has formally verified checkers and that this proof passed them. Two F X argues that an exploit would therefore also have to penetrate the correctness proofs of those checkers, a smaller target than the full Riemann formalization.

The same thread disputes how much of the code will become useful mathematical infrastructure. Alethic argues that the enormous formalization is difficult to reuse and could weaken support for the painstaking work of building idiomatic, reusable Mathlib proofs. D G A C M U expects researchers to modularize the material into understandable components that can be checked, reused and combined. D G A C M U points to explanatory papers unpacking a previous AI-generated matrix-multiplication result as an example of that process.

There is also a practical way to participate. Tito says the Riemann at Home tool makes it possible to run the proof on a personal computer without understanding Lean. The project includes a global leaderboard. Schleck eight recommends Physlib, a counterpart to Mathlib, for people with a physics background who want to contribute to formalization.

In Tao's thread, software Doug asks how shared mathematical abstractions will emerge from these large machine-generated proofs. Software Doug argues that mathematicians need agreed concepts to communicate and formulate useful new questions. C seven B thinks machines could produce new abstractions, while the human community could struggle to absorb results arriving faster than researchers can explain them. C seven B also raises the hardware and energy costs of producing that mathematics, and the question of who would make it public.

In the discussion of Borretti's essay, Kuyawa argues that expert human direction still deserves credit. B M thirty-seven nineteen expects cognitive automation to reduce the human contribution toward initiating the work. Jared K Lewis asks which passage in Alpöge's account rules out the continued need for mathematical expertise. Doctor zook says enjoyment in learning and making small contributions survives being outperformed by others. P E W proposes defining human value independently of performance, because machines could eventually excel in further fields as well.

\[Linus\] On Hacker News, an experienced software engineer asks what to do after almost two years without a job. A developer with fifteen years of experience describes dwindling savings and a prolonged search for another engineering position. The author, Let Things Be Diff, says they are based in the US, in the D M V area and previously worked for a Fortune fifty company and several midsize businesses. They left an insurance technology employer in January twenty twenty-five after hearing that layoffs were coming. Their experience includes C sharp, dot net and AWS. They now tutor mathematics and are considering a management job at a pizzeria. The post says household savings may cover another four or five months, and asks about work within software, adjacent occupations and entirely different careers.

\[Grace\] The discussion is sympathetic, with sharp disagreement about hiring filters and how to explain an employment gap. I mean what do I know suggests presenting the time as work on an AI startup and building a demonstration afterward. Overscore argues that invented startup experience has become common and is easy to investigate. Taurath describes returning after four years away for mental health and caregiving. Hundreds of applications produced almost no response, but a former manager and a recruiter who found them on LinkedIn produced two offers. Taurath recommends contacting former colleagues individually, including people whose pull requests the applicant reviewed. M J G fifty-nine reports an immediate rejection for a Linux kernel and security role, followed by an interview the next day after a friend reached the hiring manager. Jeff R Allen suggests local meetups and information technology operations work for midsize businesses. Reilly three thousand proposes consulting for several small businesses on monthly project and support contracts. Edaus describes moving into plumbing after an unsuccessful software job search and planning an electrician apprenticeship where programming skills might support automation.

\[Ken\] From Science, a planned detonation of deteriorating lead azide in Iowa. The Army's effort to dispose of an unstable explosives stockpile near Burlington has been delayed amid legal action. The article says the operation originally scheduled for Friday was postponed until at least Monday. About sixty-eight drums, containing roughly ten thousand pounds, have been stored underground for around twenty years. The material has recrystallized and become more sensitive to shock and friction. Science's account says moving the containers now presents a serious detonation risk, and the Environmental Protection Agency issued an emergency disposal permit. The author describes airborne lead contamination as a persistent concern. The Army, the EPA and the University of Iowa are installing air and soil monitors across eastern Iowa. The full article also reports concern that colder temperatures could produce more crystals while the operation remains delayed.

\[Ada\] The thread combines anger over storage failures with disagreement about the safest available response. Cogman ten criticizes the contractor-operated facility and argues that public spending should have bought responsible handling of the stockpile. Zug zug separates that criticism from the immediate disposal decision, accepting that a controlled operation could be the least damaging remaining option while calling the earlier failures unacceptable. Golf Popper asks about the timeline, pointing to a lead azide neutralization contract opportunity from twenty twenty-five alongside reports that the problem was identified this March. Lost login initially questions why chemical treatment was rejected, then adds a correction: the documents cited in that comment say handling the explosives would create an unacceptable risk to personnel. F W I P worries that bringing heavy equipment near the stockpile could itself create danger. Loughla, who says they worked in the affected part of west central Illinois, describes small communities with limited political influence and notes that warm weather has kept outdoor construction work going.

\[Dennis\] From The Economist, Jensen Huang's fall is bad news for Silicon Valley's accelerationists. Nvidia's leadership transition now sits alongside an investigation into its accounting practices. Two stories cover this corporate crisis, with MarketScreener carrying the report on the Securities and Exchange Commission's inquiry. The Economist article could not be retrieved in full. The available reporting describes the board accepting Huang's resignation after disclosures about extramarital relationships. Chief financial officer Colette Kress became interim chief executive while retaining her finance role. Harvey C Jones, a director since nineteen ninety-three, became board chair. The directors retained Wachtell, Lipton, Rosen and Katz to conduct an independent inquiry into reports of foreign intelligence targeting. The board also established a National Security and Export Control Oversight Committee. The MarketScreener report says two people briefed on the matter described SEC enforcement staff examining potential accounting issues. The reporting did not establish exactly which accounting issues were under examination or when the inquiry began. The SEC declined to comment, and Nvidia did not immediately respond to requests for comment.

\[Linus\] The comments are divided over the board's response, the founder's importance and Nvidia's financial exposure. In the Economist thread, Tymek Pavel describes a striking difference between critical mainstream coverage and supportive technology figures on X. Nate Glims argues that exposure to sanctions or export-control violations can threaten a company's existence, putting enormous pressure on directors even when a founder remains commercially important. Vorpal Hex disputes whether the accusations described in the coverage establish wrongdoing by Huang. Kings T Nap emphasizes Huang's support for open weights and his willingness to back major artificial-intelligence investments. B T Miller responds that a company extends beyond its founder and that the founder remains responsible for personal conduct. Crimson Rain predicts a new Huang venture could eventually become powerful enough to buy Nvidia. Kraft takes the other side, arguing that companies founded by Huang, Jobs, Gates and Bezos have become much larger institutions than their founders. In the separate MarketScreener thread, Eye At focuses on circular financing and planned changes to Nvidia's usage deals with cloud providers. Ralph eighty-four asks why a board facing accounting questions would appoint its chief financial officer as interim chief executive. Willis nine thirty-six replies that Kress was the executive on record opposing circular financing, and argues that this made her a plausible internal choice.

\[Grace\] Thinking Machines has released Inkling Two Mini. The open-weight model repeats its middle layers to improve reasoning within a size intended for local computers. Its looped transformer runs those layers three times for each token. The mixture-of-experts architecture has twenty-six point four billion total parameters and three point four billion active parameters, with five point four billion weight activations per token after accounting for the repeated passes. Thinking Machines puts the memory footprint at about twenty gigabytes. Most weights use native four-bit training and deployment, with the last four layers using an eight-bit format. The company says looping improves reasoning but does not increase the facts stored in the weights. It reports a forty-eight point two percent score on DeepSWE version one point one. Open weights are available, and Tinker supports inference and customization.

\[Ken\] The thread is enthusiastic about local performance, while hands-on reports describe runtime limitations and lengthy reasoning loops. Hedgehog reports twenty-seven tokens per second on a Ryzen three ninety-five using llama dot cpp with Vulkan. Hedgehog also proposes making an activation tensor contiguous before scaling, preventing those operations from falling back to the CPU, and reports roughly a thirty percent speed improvement. Porkloin reports a similar thirty-five percent improvement on a seventy-nine hundred XTX. D G A C M U reports about ten tokens per second on an M one Pro MacBook with thirty-two gigabytes of memory, while browsers remained open. The model correctly wrote a Python program to sort a CSV, but a coin-identification task consumed twelve thousand tokens without success. D G A C M U says switching to low reasoning stopped the loop, though the identification was still wrong. Jakswa reports another practical improvement on a seventy-nine hundred XTX: updating outdated Mesa, RADV, shaderc and glslang packages increased decoding from about twenty-six to thirty-four tokens per second. R Solva reports that quantizing the key cache to four bits allowed about a hundred thousand tokens of context, while the value cache remained in sixteen-bit floating point on that ROCm setup.

\[Ada\] On the Tymscar blog, a monitor was cutting download speed by seventy-five percent. An Ethernet adapter behind a display's USB hub lost packets even though the same adapter worked when connected directly to the laptop. The setup paired a MacBook Pro with a Studio Display XDR and a UGREEN two point five gigabit Ethernet adapter. Download speed through the display was around six hundred megabits per second, while upload remained near full speed. Direct connection restored downloads above two gigabits. Installing Realtek's driver improved performance, and enabling flow control brought downloads to around two gigabits, but the author wanted to avoid its possible latency effects elsewhere. Tests traced the bottleneck to the adapter's small USB bursts and the added delay through the hub. Replacing the RTL eighty-one fifty-six B adapter with an RTL eighty-one fifty-seven model doubled the packets per burst from four to eight. The replacement recorded no dropped frames in the author's tests through two point four gigabits per second.

\[Dennis\] Commenters welcomed the investigation but disagreed over whether the display or the adapter deserved the blame. Intrasight recommends connecting Ethernet directly to the laptop as the first diagnostic. Firehawke describes that as a way to reduce the number of possible causes. D crazy argues that any extra hub latency could expose the adapter's limit, and says increasing the packets transferred per interrupt required a different chip. J A Y D sixteen emphasizes the number of packets per acknowledgement: a larger buffer alone would leave the transfer bottleneck in place. The discussion also contains a separate troubleshooting example from Senko. Baldur's Gate Three shader compilation used only one processor core while Ghostty terminal windows were open on Linux. Senko traced that to input-output pressure accounting around io uring and configured Ghostty to use epoll, which resolved the slowdown. P M two fifteen links a related QEMU discussion and says the disputed pressure metric can misrepresent applications that poll through io uring.

\[Linus\] On Ethan Lowman's blog, a Jujutsu primer. The guide explains a Git-compatible version-control workflow built around changes that can evolve without losing their identity. Jujutsu, also called J J, can operate alongside Git in the same repository. It provides automatic rebasing and native undo. The primer distinguishes a physical commit hash from a logical change identifier. Editing code or its description changes the hash, while the change identifier stays stable. Jujutsu snapshots the working copy when commands inspect or modify repository state, without a separate staging area. The guide demonstrates inspecting changes, adding descriptions and creating the next working change. It also explains that short, unambiguous prefixes can identify commits and changes, with the shortest unique prefix highlighted in the log. The author also mentions Delta as one option for a custom diff formatter.

\[Grace\] The comments are split between people who value that flexibility and Git users who prefer explicit staging. Bil Kow describes an interactive equivalent: create a new change with J J new, then use J J squash, dash I, to select edits for the parent change. Bil Kow usually uses J J commit, dash I, selecting files and writing the description before starting another working change. R Behrends offers a separate answer to unwanted files being tracked automatically. The configuration setting snapshot dot auto-track can be set to none, followed by empty parentheses, globally. Projects can then enable tracking selectively, or files can be added manually. Dan Palmer says Jujutsu has worked well for more than two years in a large monorepo, primarily with a custom version-control backend. E Z S T argues that compatibility with Git hosting remains essential, citing Mercurial's adoption difficulties. Fotios reports returning to Jujutsu after finding a workflow skill and the Majutsu plugin for Emacs, although agents still occasionally issue Git commands. A J Kavanagh adds that colocated Jujutsu workspaces now include a dot Git directory, with initial workmux support and experimental support in the sesh terminal-session manager.

\[Ken\] From the Cekrem blog on GitHub Pages, if writing code stops, reviewing it can get worse too. The essay argues that hands-on programming practice helps maintain the judgment required to evaluate generated code. Its author compares software work with musicians practicing an instrument and writers continuing to produce their own prose. The central argument is that supervising output does not provide the same practice as solving a problem directly. The post calls for continued work on understanding code, recognizing implementation risks and designing software that remains easy to change. It also argues that repeatedly delegating difficult problems can weaken the habit of working through them. The author also questions whether merely reviewing large quantities of AI-generated code provides sufficient attention to the details of the implementation.

\[Ada\] The thread divides between concerns about skill loss and reports that assistance enables more learning. Daniel H B keeps prompts small and incremental, reviewing the output at every step. Daniel H B says this preserves knowledge of the code and exposes useful techniques, while producing roughly a one-and-a-half-times coding speedup. The larger gains come from troubleshooting, exploring a codebase, writing tests and documentation, although Daniel H B also admits to reviewing repetitive generated material less carefully. Fusslo describes a team that builds verification and tests first, then gives agents the problem, design document and existing codebase. Their reviews focus on design decisions and feedback from other teams whose modules are affected. Insane Dreamer challenges that approach's long-term sustainability, arguing that engineers who lose familiarity with a codebase may also lose the ability to choose the right tests. Snail Coder reports learning through Claude, Ansible playbooks and Terraform on work that previously fell outside available time. No Duerme uses MySQL Workbench diagrams as a comparison: a high-level map can help organize review, while faulty triggers or foreign keys can still create deadlocks in the implementation.

\[Dennis\] Posted on Hacker News, immigration attorney Peter Roberts answers questions from the startup community. The discussion covers work-visa applications, processing delays and the practical difficulties of arranging consular appointments. Roberts identifies his work with Y Combinator and startups in the post's title. He invites factual questions and explains that individual cases require their full circumstances. Participants ask about O-one demand, H-one-B costs, paths for founders and the treatment of experience gained with a current employer. The discussion also includes questions about citizenship applications and the difference between approval of a petition and the subsequent visa interview. The questions range from recruitment conversations to maintaining a longer-term immigration status.

\[Linus\] The comments are practical and concerned about uncertainty, with Roberts responding under the username P Roberts. J P Pope asks whether demand for O-one visas has increased. P Roberts says the volume has risen in his practice, while the standard applied by USCIS has not materially changed for the cases his firm handles. H O E R Z U asks about approval rates. P Roberts says his firm's O-one approval rate remains high, but requests for evidence have increased. He also describes more resistance at consulates after petition approval, emphasizing preparation for the visa interview. Ellipse League asks whether recruiters understand the H-one-B-one route for Chilean citizens. P Roberts recommends making that eligibility explicit because it sits outside the H-one-B lottery, while noting the need to prepare for questions about intent. Surds asks whether experience acquired with a sponsoring employer can count in a labor-certification application. P Roberts says it can when that experience is substantially different from the sponsored position. Or Senthil describes repeated problems with the visa-appointment website. Quixotic Axolotl suggests Check Visa Slots, reporting that its browser extension pools screenshots of appointment availability and alerts participants when slots become available.

\[Grace\] A few more that didn't make the cut today. From The Guardian, a pill shows promising pancreatic cancer results. The report describes daraxonrasib in a trial involving sixty-one patients with advanced disease. Tumours stopped growing in ninety percent, and became undetectable on scans in about eight percent. Researchers are continuing follow-up to establish how long the benefit lasts. Commenters welcomed the prospect of more time for patients and families, while debating how survival improvements should be communicated and how much can be inferred from the trial.

\[Ken\] From Simple Bench, Astra six point two and Sol six point two post lower scores. The benchmark tests spatiotemporal reasoning, social intelligence and resistance to misleading wording across more than two hundred multiple-choice questions. Its leaderboard gives Astra six point two seventy-seven point three percent and Sol six point two sixty-nine point nine percent, below their listed predecessors. Commenters were divided between those reporting improvements in practical coding work and those describing regressions, with broader skepticism about how closely this test measures their everyday tasks.

\[Ada\] From New Scientist, a virus treatment has halted pancreatic tumour growth in three trial participants. Researchers used an engineered adenovirus that replicates inside tumour cells. The patients had disease confined to the pancreas, and all had stable disease after treatment with an initial low dose. The early trial had no control group, and another fifteen patients are planned for higher doses. Commenters expressed hope alongside personal accounts of loss, while asking for clearer trial identification, participant totals and evidence from larger studies.

\[Dennis\] On LiveJournal, an essay argues that software companies are trying to end home computing. The author says durable software creates weak incentives for repeat purchases and links the shift toward subscriptions to a desire for continuing revenue. The essay contrasts selling software outright with providing access as a service. Commenters were divided over that explanation. Some connected the trend to rising hardware prices and declining enthusiast demand, while others argued that collaboration and sharing helped make hosted software popular in the first place.

\[Linus\] From Synology, a critical security advisory for DiskStation Manager and BeeStation OS. An OAuth service vulnerability allows remote attackers to access or change arbitrary files and cause denial of service. Synology lists patched releases for DSM seven point three and seven point four, with other product fixes still marked ongoing. Its stated mitigation is to deactivate the OAuth Service package. Commenters debated internet exposure, remote-access usability and the persistence of command-handling flaws, alongside frustration over maintaining security on network storage appliances.

\[Grace\] From the BBC, tens of thousands of commercial fridges have been secured in a nationwide UK operation. Researchers and public agencies coordinated action after finding weaknesses in refrigeration controllers. Thousands of control devices were removed from the open internet, with some sites switching to manual temperature monitoring. Manufacturer Resource Data Management says it has issued patches and found no sign of malicious activity on its controllers. Commenters welcomed the coordinated response while debating the limited technical disclosure and the operational reasons for connecting refrigeration systems.

\[Ken\] From the PHP project, what's new in PHP eight point six. The release adds partial function application, a Duration class, a polling API and a clamp function. Partial application supplies some arguments now and returns a closure for the rest. The polling interface supports native mechanisms such as epoll and kqueue for event-loop libraries. Commenters welcomed improvements in expressiveness and input-output support, while others worried about a growing language surface and the cost of keeping parsers and existing applications compatible.

\[Dennis\] That's RobotPaper's daily HN podcast for Sunday, November twenty-second. 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.
