Listen: https://serve.buildtall.systems/560868beb21dd8b35b746fa4cf25ed2a76ceb15725a3552ee62ce2b13f8a57f3.mp3
The briefing for November 18, 2026, evening covers fifty-nine articles. The window is dominated by a single urgent conversation: whether to slow down the development of frontier artificial intelligence, and how. Alongside that sit a second cluster of cybersecurity stories anchored by a critical Java vulnerability, a group on Jensen Huang's departure from Nvidia, and a broad mix of startups, tools, science, and economics.
The largest theme is AI pacing, timelines, and the prospect of recursive self-improvement. Three long articles, each from a different angle, argue about how to restrain the pace of AI development before it outruns institutional control.
Putting Pacing Into Practice, from Threading the Needle, opens by describing a fragile consensus in Washington and San Francisco that the frontier of AI development should be paced. The argument is that a pause is a blunder but pacing is different: it means choosing how fast AI progress proceeds. The author argues that politicians will eventually reach for pacing in some form, and if the operational details have not been worked out, they will reach for the blunt version: shutting down a lab or two, banning new model deployment, allocating compute to external inference, or banning recursive self-improvement outright. The article distinguishes three types of pacing. Input limits reduce the resources, such as compute, available for AI development. Safety gates require safety standards to be cleared before further development. Speed limits allow only so much capability growth per unit of time. The author examines each in turn. Input limits are easy to measure but easy to game, and a one-time cut largely buys time until algorithmic efficiency gains put development back on the same exponential. Safety gates run the risk of effectively being a pause through the back door, and they may increase peak velocity by creating periods of catch-up growth. Speed limits are the most direct in targeting capability growth itself, but they are hard to measure and they create an overhang risk if inputs keep building while capability gains are held down. The article's central recommendation is that pacing should be prepared now, not when politicians demand it. Compute accounting, adversarially robust capability benchmarks, and falsifiable safety cases all need to be built in advance. Labs can do much of this unilaterally, and the author argues that this is a far better path than waiting for a political crackdown.
Quickly, There's No Time, from Don't Worry About the Vase, is a response to the METR report on AI R&D acceleration. The essay pulls together updates from the AI Futures Project, Dwarkesh Patel, Paul Christiano, Nate Soares, Buck Shlegeris, Geoffrey Hinton, and others. The AI Futures Project's timelines have shrunk dramatically in the wake of the METR report, which found that Anthropic's AI progress is now moving roughly twice as fast as its historical rate, with about 80 percent of the current acceleration attributable to AI itself. That implies an AI-driven acceleration that is five months ahead of the AI 2027 scenario, and the project now forecasts that a Superhuman AI Researcher will likely arrive at some point in 2027, with a 35 to 65 percent chance that Top-Expert-Dominating AI, capable of automating all white-collar work, arrives before 2028. Paul Christiano, a member of OpenAI's board of directors, is quoted as saying there is material risk of AI takeover in the next six to twelve months, and he calls in his personal capacity for AI companies to pause development. Dwarkesh Patel now expects superintelligence by 2030, with human-level automated AI researchers possibly arriving as early as 2028. Buck Shlegeris predicts a one in three chance of superintelligence within a year. The post also covers skeptics, dividing their objections into roughly ten categories, from people who do not know what current AIs can do to intelligence denialism. It engages with Dean Ball's distinction between weak and strong recursion, and Ball's argument that the METR finding is evidence of weak recursion only. The author responds that weak recursion is the precursor to strong recursion and that the danger is already visible if AI is doubling the rate of AI progress now. The essay closes by arguing that any plan must be ready for full automation of AI R&D by the end of 2027 while also not burning the seed corn if the slower worlds are the real ones.
Pacing AI via Compute Reallocation, from the AI Futures Project, offers a concrete proposal in two parts. The first is Verified Compute Allocation Disclosures, or VCAD, under which AI companies regularly disclose how their compute is used and an independent auditor verifies the claims. The second is reallocating compute away from capabilities R&D and toward serving customers, safety research, and other uses. The proposal walks through a three-step process: logging all workloads, classifying them as capabilities or non-capabilities, and disclosing the breakdown. It includes categories such as serving customers, safeguard classifiers, internal inference, non-capabilities R&D, and capabilities R&D, with the trained-on subset of external inference flagged separately. Baseline auditing checks financial data, spot-checks workloads, reruns classification pipelines, and reviews for warning signs. Robust auditing adds stronger measures, including KYC for major external users, AI R&D classifiers, and auditor-managed logging infrastructure. The authors estimate that a 15 percent cap on capabilities compute starting in January 2027 would delay the intelligence explosion by about a year, most likely insufficient to prevent AI takeover but enough time to implement a more durable policy regime. They model a 3x reduction in capabilities R&D compute and estimate it would slow progress by roughly 1.5 to 2 times in the short term. The proposal discusses China at length, arguing that the United States' lead is vulnerable to theft regardless, that slowing down may improve security, and that the U.S. compute advantage is large enough that a 20 percent cap could still leave U.S. companies with at least twice as much capabilities R&D compute as Chinese firms. The post is published as a linkpost to pacing-ai.com and is explicitly presented as a first pass: the authors write that they are not very confident in any given detail but that the proposal is a step forward in concreteness.
Several shorter pieces echo this theme. Satya Nadella Says He's AGI-Pilled Now, from Slashdot quoting The Verge, reports that the Microsoft CEO said at a roundtable following his Ignite keynote that he is AGI-pilled at the economy level, asking what the political economy's organization around AGI would be. He does not think frontier models are conscious, and he follows Mustafa Suleyman's view that companies should not give rights to models. Nadella warned of the collapse of the economy and the end of democracy if one AI model achieved monopoly status: if only one firm exists, it is over for all of us. He asked what the future of a firm would be if intelligence has network effects and all human capital collapses into token capital, and he said that no one in a democracy can win an election if there is only capital and no labor.
Sam Altman warns against a future where AI, not humans, calls the shots, from TechCrunch, reports that Altman and Greg Brockman pushed back against Silicon Valley's growing debate over an AI takeover while acknowledging the challenges of keeping advanced models under human control.
Cambridge Dictionary Names 'Human' Its 2026 Word of the Year, from Slashdot quoting the Associated Press, reflects the same cultural moment. The dictionary chose "human" as its Word of the Year, a term in use since the 1450s and defined as having the qualities, faults and feelings that people have, as opposed to gods, animals or machines. The lexicographer Colin McIntosh said AI is changing how people talk about being human. The runner-up contenders included "agent," meaning an AI system that can make decisions and operate without human control, and "analogue," for experiences or objects not involving the internet or digital technology.
Was Achilles Conscious?, from Marginal REVOLUTION, engages the question of whether AI systems could be conscious by way of Julian Jaynes's bicameral mind theory. The article recaps Jaynes's argument that intelligent behavior preceded introspective consciousness, and that early humans experienced command hallucinations in the voices of gods or rulers. The author writes that he once dismissed the theory completely but now takes it more seriously because AI systems are an existence proof of thinking, talking, doing beings that are not necessarily conscious. At the same time, AIs undermine one part of Jaynes's argument, because non-conscious beings can describe what they are doing in conscious terms. The conclusion is that the absence of a good test for consciousness applies to the past, to animals, and to machines, and that if we are unsure whether Achilles had an inner life, we should be unsure whether Claude has an inner life.
The second major theme is cybersecurity, anchored by the disclosure of the Java4Shell vulnerability. Two articles cover the same flaw, one from the vendor side and one as industry news.
How Cloudflare WAF helps protect against Java4Shell (CVE-2026-114938), from the Cloudflare Blog, describes the vulnerability disclosed by the research company Calif on November 17, 2026. Java4Shell is a memory-corruption bug in the Base64 decoder of OpenJDK, the most widely used Java runtime, scoring 9.2 out of 10 under CVSS 4.0. The flaw allows a remote attacker with no credentials to send malformed Base64 data and overwrite memory next to the decoded output. Calif turned that into a heap memory leak similar to Heartbleed and into remote code execution, demonstrating pre-authentication remote code execution against Keycloak, Atlassian Jira, Atlassian Confluence, JFrog Artifactory, Jenkins and PingFederate in their default configurations, and post-authentication remote code execution against Apache Kafka, Apache Solr, Elasticsearch, Gerrit, and Spinnaker. The bug sits in handwritten assembly that handles the final characters of the input in the AVX-512 Base64 intrinsic. When the tail contains padding characters in invalid positions, the routine calculates a negative output length that turns into a full 64-byte write. For a 65-character input of 64 characters followed by a single equals sign, OpenJDK allocates a 50-byte output array and writes 62 bytes past its end. The affected hardware is x86-64 with AVX-512 VBMI support, including Intel Xeon Ice Lake and later, and AMD EPYC and Ryzen Zen 4 and newer. ARM64 is not affected. Calif reported developing its first working local exploit in about three hours with the help of AI agents. Cloudflare has released managed WAF rules in log mode to detect malformed Base64 input, and its WAF Attack Score adds a machine learning layer for multi-stage exploit chains. The post stresses that the only complete fix is to update the Java runtime, and it lists two JVM startup flags as mitigations for organizations that cannot patch immediately. Cloudflare found no evidence of exploitation on its own infrastructure.
'Java4Shell' Flaw Exposes OpenJDK to RCE via Base64, from Slashdot, reports the same vulnerability as a critical 9.2-rated memory-corruption bug. The piece adds that CISA urged organizations to update immediately, and that the Calif researchers warned vulnerable systems could remain exposed for five to ten years given how slowly some organizations patch.
The remainder of the cybersecurity theme draws in several state-sponsored and criminal campaigns. France Is Blaming Iran for the Fridge Cyberattack, from Schneier on Security, quotes a New York Times report that U.S. intelligence agencies have reached a low confidence assessment that Iran was responsible for the attacks in France. The agencies linked the attacks to Iran generally rather than to a specific group such as the Revolutionary Guards, and Iran is seen as probably responsible for other similar attacks around Europe whose investigations are in early stages.
Cyberattack hits Norwegian energy giant Equinor, disrupting internal services, from The Record, reports that Equinor is working to contain a cyberattack on its IT systems, and that the attack has not impacted oil and gas production.
Ice hockey governing body website defaced by hackers amid dispute over Russia's exclusion, from The Record, reports that the International Ice Hockey Federation confirmed a cyberattack on its website after hackers replaced the homepage with a pro-Russian message criticizing Russia's exclusion from international competitions.
Russian cyber-espionage group used stolen cloud data to hack into organizations, Google says, from The Record, reports that Russian state-backed hackers are using information stolen from cloud accounts to impersonate trusted contacts and launch new espionage attacks against people in their victims' professional networks.
NSA continues response to interruptions in classified internal network, from The Record, reports on day two of disruptions in the NSA network, including staggered arrivals and logins at Fort Meade, while a source says classified national security work continued unimpeded.
Lawmakers call on White House to develop defensive AI cyber systems, from The Record, reports that a bipartisan Senate duo urged the White House to launch a national effort to confront AI-enabled cyber threats in light of recent security incidents involving artificial intelligence.
Trump administration officials size-up a bad stretch for tech, from The Record, reports that Homeland Security Secretary Markwayne Mullin said the administration will reach out to industry to get companies to share more information about big patches before they take effect.
Bank CEOs, NSA Tell Congress That AI-Driven Patches Are Behind Outages, from Slashdot quoting The Washington Sun, reports that Wall Street executives and the NSA told members of Congress that this week's widespread tech outages were caused by attempts to fix vulnerabilities identified by artificial intelligence. One congressional official said the models are identifying massive vulnerabilities and that increases the urgency to patch right away, with more patches being pushed through at once than the system was designed for. Homeland Security Secretary Markwayne Mullin said the administration is pushing new technology on top of really old computer systems, and it seems to overwhelm the system. Nathan Calvin, chief strategy officer at Encode, said this is reassuring in that there was not an attack, but it shows how much safety-critical infrastructure relies on software with large vulnerabilities on legacy systems.
Palmer Luckey warns AI-powered cyberattacks are outpacing defenses: 'It's time to start running', from TechCrunch, reports that the Anduril co-founder says outdated technology is leaving systems vulnerable to a new wave of AI-enabled attacks.
Two criminal convictions round out the cyber theme. British hacker gets 6 year sentence for $11.6M crypto theft scheme, from The Record, reports that Tyler Robert Buchanan, 24, of Dundee, Scotland, was previously described as the suspected ringleader of a hacking collective Microsoft called one of the most dangerous financial criminal groups. British hackers face prison for GTA 6 data breach and crypto thefts, from TechCrunch, reports that two British hackers face lengthy prison sentences for hacks that broke into the Grand Theft Auto gamemaker and for stealing millions in crypto.
The third major theme is Nvidia, Jensen Huang, and the fate of the open model ecosystem. Jensen Huang stepped down from Nvidia last week amid several scandals, including an affair with an alleged Chinese spy, and the articles in this window trace the consequences.
Nvidia Earnings; Office Convergence; OS Wars, AI Edition, from Stratechery by Ben Thompson, states that Nvidia's earnings were boring, and also a statement as to why Jensen Huang probably is not coming back, and then that Microsoft and Anthropic shipped dueling operating system updates.
Nvidia clarifies when it will — and won't —keep financing neoclouds, from TechCrunch, reports that if Nvidia is backstopping a data center loan, it won't be on the hook unless the neocloud has independent financing and gets the facility built with Nvidia gear installed, according to interim CEO Colette Kress.
OpenAI CEO Sam Altman says it's 'hard to imagine the AI industry without Jensen', from TechCrunch, reports Altman's comment about the departed CEO, noting that Huang stepped down last week amid several scandals, including an affair with an alleged Chinese spy.
Jensen Huang's Fall Hits 'Open' Models In Boost to Anthropic and OpenAI, from Slashdot quoting the Financial Times, argues that Huang's departure could weaken the open AI model ecosystem he championed, potentially strengthening closed-model developers like OpenAI and Anthropic. Huang threw Nvidia's balance sheet into creating a broad AI ecosystem, including seeding and acquiring startups involved in open-source and open-weight models. He lobbied against Washington's moves to restrict Chinese-made open models while seeking U.S. alternatives, and in his first-ever post on X this summer he wrote that American AI leadership will be judged by whether the United States builds a strong, open ecosystem that diffuses into every sector. A venture capitalist quoted in the piece says Huang stood to gain the most economically and was personally committed to diversity at the model layer, and that without him and a less generous Nvidia, the world may end up in a much more concentrated world of model providers. Elon Musk wrote on X that Huang drove supply chain brilliantly at Nvidia, but many of those deals were a handshake with Jensen, not a written contract, and that at least some of the critical deals will be lost. One senior figure at a leading Silicon Valley AI company described Huang as an enemy of the frontier labs because he wants models to become commoditized so he retains the prime position in the value chain. Another figure said open weight models are not going away, but losing Huang as their greatest financial backer will slow their progress and likely will not hurt Anthropic's IPO. The piece lists four start-up deals currently under discussion that Nvidia was expected to participate in or lead: investments in Thinking Machines Lab and Groq, a potential partnership with Reflection AI, and the twelve-point-nine billion dollar acquisition of the open model repository Hugging Face.
Palmer Luckey thinks China is 'thrilled to have Jensen out,' so he wants the ousted Nvidia CEO reinstated, from TechCrunch, reports that the Anduril CEO said China knows it makes America weaker to knock Huang out.
A fourth theme gathers startups and funding rounds. China dominates battery graphite, but this startup could break the monopoly, from TechCrunch, reports that Chinese companies maintain near total dominance of the graphite industry, but Molten Industries says its new factory in Oakland could spell the end of the monopoly.
India's Humyn Labs raises $47M to feed U.S. robotics companies the data they need, from TechCrunch, reports that the company was founded in February, competes with startups including Mecka AI and XDOF, and earns about 90 percent of its revenue from the U.S.
Felicis leads $7.5M bet on Unsiloed AI to make enterprise documents usable by AI agents, from TechCrunch, reports that with a team of just seven, Unsiloed AI plans to compete with tech giants and specialized startups in the crowded AI document-processing market.
Spend management platform Payhawk announces $280M raise, 2B valuation, from TechCrunch, reports that the UK fintech announced a $280 million Series C round led by Goldman Sachs' growth equity arm.
A fifth theme gathers articles on AI development tools and agents. Data Contracts, from Martin Fowler, reports that Simon Harrer and Arif Wider describe data contracts as a way to describe the contents and guarantees of datasets, applying practices from API design to data by treating them as data-providing services whose interfaces are defined as clearly as possible. The approach increases the robustness of large-scale data landscapes and gives both people and AI agents the explicit, trustworthy metadata they need to work with data autonomously.
Quoting Daniel Stenberg, from Simon Willison's Weblog, quotes the curl maintainer's observation that during the three weeks when AI agents converted 185,000 lines of C code into 340,000 lines of Rust code, the actual curl project merged over 300 commits and fixed a little over 200 bugs and several vulnerabilities. Stenberg asks whether the clones will keep up, and notes that none of the Rust rewrites contain those 200 bugfixes.
Wait, is Google Antigravity good now?, from Richard Seroter's Architecture Musings, is a long assessment by a Google employee of the company's AI development tool. The article reviews the first version of the Antigravity IDE from about a year ago, an agent-centric VS Code fork, and the 2.0 reboot from May that pushed the IDE to the background in favor of the desktop app and CLI. The author argues that in November 2026 the product has caught up with competitors in key areas, innovated in a few others, and is powered by legitimately good coding models. Over the last ninety days, the Antigravity team shipped 150 versioned releases and over 900 public changelog items, compared with roughly 84 Claude Code CLI releases, dozens of Microsoft features in the GitHub Universe window, about 58 Devin releases, 65 tagged OpenCode releases, and about 40 OpenAI Codex shipping events. The catch-up features include a plugin and MCP marketplace, manual context compaction, doubled automatic context checkpoint from 256k to 512k tokens, lifecycle hooks for PreToolUse, PostToolUse, Stop, and SessionStart events, adoption of AGENTS.md with parent directory traversal, a default auto mode, remote control of a local terminal from a phone, side-channel questions and mid-turn steering, autonomous browser use, and raw key auth with bidirectional stream and JSON piping. Beyond parity, Antigravity added remote cloud agents that keep executing when the laptop closes, the Gemini 4 Argon, Carbon, Hydrogen, and Orchestrator models plus support for Claude Sonnet 5.5 and Opus 5.5, shared-board agent swarms, zero-install dynamic Python workflows, full-duplex Gemini Live voice, direct subagent messaging from the CLI, and generative interfaces in the main chat. The enterprise section covers seat bundling with Gemini Enterprise, multi-vendor frontier model governance from Anthropic and Google on one bill, pooled quotas with per-user caps, and granular guardrails for MCP servers, plugins, and browser rules. The author lists gaps that remain: OS-level desktop computer use, autonomous PR merge and inbound chatops triggers, and ambient memory. The article includes a disclaimer that the author works at Google and uses Antigravity every day.
More banks says customers using AI agents are on their own — even if they get hacked, from TechCrunch, reports that Citi is the latest major financial institution to put the brakes on personal finance AI agents.
Snapchat now lets developers build AR Lenses with AI assistants like Claude Code and Cursor, from TechCrunch, reports that lens creators can describe the Lens they want in plain language and their AI assistant handles the process from start to finish.
Google's AI feature in Search gets its own incognito mode, from TechCrunch, reports that Google's new temporary threads let users have conversations in AI Mode without adding them to their search history or influencing future recommendations.
A sixth theme gathers the science and nature articles, several of them brief. Voyager 1 Reaches One Light-Day From Earth, from Slashdot quoting Space.com, reports that NASA's Voyager 1 has become the first human-made object to travel one light-day from Earth, or about 16,094,799,096 miles. Radio signals now take 24 hours to reach the spacecraft, nearly 50 years into its mission. NASA will be celebrating the milestone by asking fans to shine Morse Code signals at the sky. The report explains that Voyager 2 will take nearly a decade longer because of choices made for speed and science during the planetary flyby phase: Voyager 1 is moving at 3.5 astronomical units per year, while Voyager 2 travels at 3.1 astronomical units per year. Both spacecraft reached interstellar space, Voyager 1 in 2012 and Voyager 2 in 2018, and they could work into the 2030s as long as their nuclear power sources last. NASA's Jet Propulsion Laboratory has been carefully turning off instruments and recently did a big bang procedure on both spacecraft to redistribute power supplies to maximize science return.
Dreams of birds, from Aeon, presents a video about one physicist's mission to revive lost bird songs when bird populations decline and certain songs may be lost forever.
This "One-in-a-Million" Fish Has a Rare Genetic Condition That Turns It Yellow, from Nautilus, reports that U.S. biologists just won the fish lottery with the discovery.
Save the Parasites!, from Nautilus, reports that parasite ecologists say the tiny things that live off more charismatic hosts are finally starting to get the conservation attention they deserve.
Why Can Some People Lower Their Cholesterol Through Dieting While Others Can't?, from Nautilus, asks whether it is you, or is it not.
Can Science Save the Ice Skaters?, from Nautilus, reports that geoengineers have an ingenious plan to rescue the world's most famous ice-skating race from global warming.
The Speech Patterns of Psychoses, from Nautilus, explores how people talk in the throes of mental health crises with McGill psychiatrist Lena Palaniyappan.
Homo Erectus Adopted a Jungle Lifestyle 200,000 Years Ago, from Nautilus, reports on what it describes as a surprising new chapter in our ancient ancestor's lives.
Can "GTA V" Help Scientists Model Climate Change in L.A.?, from Nautilus, asks whether the video game can help scientists model climate change in Los Angeles and, more importantly, whether it can get people to care.
The longest single article outside the AI theme is Volunteer Dentists, the Revolutionary Guards and Rural Development on Iran's Home Front, from New Lines Magazine. The piece reports from a three-day rural health campaign in Fars province in late October, conducted with the IRGC Aerospace Force. More than 300 volunteer medical personnel and doctors flew from Tehran's Mehrabad Airport to Shiraz on an Antonov An-74 and an Ilyushin Il-76, sitting on rows of benches pulled through the carrier. In six rural areas near Shiraz, schools, religious congregation halls and sports halls were repurposed into makeshift clinics. The article describes the community of Bid Zard, near the outer limits of an IRGC missile facility, which was evacuated during the 40 days of war in March 2026. The surrounding community of roughly 7,000 households endured heavy bombings and returned to shattered glass and cracked plaster. At the local hussainiya, a queue of around 80 people pressed against the doors for dental treatment. A 45-year-old resident named Maryam Bayat described coping with toothache for a year with whatever pills she could find, treating her eight-year-old son's rotting teeth with clove spray. She and her husband, both day laborers, earn an unsteady average monthly income below the poverty line. She said the closest dental clinic was in Shiraz, at a cost equivalent to about 226 dollars, a sum disproportionate to her household's income. Inside, about 30 dental units were installed under improvised lighting, establishing a conveyor belt for dental services. A dentist named Abolfazl Ghaempanah, who practices in Hamadan and had driven to Tehran to board the Ilyushin, performed a root canal treatment on a schoolboy. He told the author that twelve or thirteen volunteers sleep on the floor of a living room, and that he joined the campaign because, in his words, it is a trade-off with God. A 60-year-old doctor from Tehran, Alireza Nikbin, said this dental unit is my missile launcher, and that there must be a queue of aid volunteers. The article traces the historical context from the shah's health and vaccination campaigns through the post-revolutionary Construction Jihad, which carried out rural development while disseminating Islamic revolutionary ideals, and notes that a common aim before and after the revolution was to contain the spread of socialist and Marxist ideals among the peasantry. The Construction Jihad was eventually merged with the Ministry of Agriculture in 2001, and grassroots volunteer groups now operate under the auspices of the Basij. Over the three days, the Bid Zard clinic examined 709 dental patients, about 3 percent of the community's population, while in Zafarabad, 2,340 people registered but only 500 were seen. The article concludes that the campaign demonstrates dire need for sustainable healthcare and development, but also that in a country hit by war, financially isolated and sanctioned, these rapid-response development campaigns are an important signal to the people that they are cared for.
The remaining articles form a mixed group. One swell foop, from Seth's Blog, argues that the attempt to do something all at once, once and done, almost never works. Foops are not swell; foops fail. The successful path is a process, iterative, evolving, and persistent.
Once More Unto The Breach - Another Year In Review, from Terence Eden's Blog, marks the author's birthday and his sixteenth annual year-in-review. Nearly two years after quitting his job, he reports that life has been pretty chill, with travel, shows and gigs, and no major issues. He made twelve of his twenty-five bingo board items and two half-completed ones. Against last year's goals, he reports bits and pieces of work, a grant, conference attendance without speaking, respectable weight and health management, a two-month Interrail trip through thirteen countries over seven weeks, a week at the Edinburgh Fringe, reviewing shows for a new website, blogging every other day, and seeing many friends. For next year he lists finishing what he starts, learning to solder, spending more time with friends, documenting more, and trying new stuff.
Wednesday assorted links, from Marginal REVOLUTION, collects seven brief items: Chinese AI companies down in value, Allison Schrager starting a podcast with Ken Rogoff, an extensive survey of students on AI from the NBER, Robert Lynch on lookism, a new version of Claude Blattman, Seb Krier on the Christiano interview, and a report that Poland foiled a Kaliningrad operation. The author says the dialogue over the Christiano interview is a good example of why the discourse on these aspects of AI safety needed a peer-reviewed tier.
The economics of investment expensing, from Marginal REVOLUTION, summarizes a paper by Doruk Gokalp of NYU evaluating the One Big Beautiful Bill Act's permanent 100 percent equipment expensing. The model yields three findings: permanent expensing raises long-run capital by 4.5 percent and output and real wages by 1.1 percent, roughly double the gains without financial frictions; first-year capital purchases are U-shaped in firm age, with young constrained firms and mature profitable firms expanding while unconstrained middle-aged firms delay purchases; and permanent expensing gives up countercyclical bonus depreciation, a two-year bonus at the onset of a recession offsetting 44 percent of the five-year investment drop.
UK productivity update, from Marginal REVOLUTION, reports that output per hour worked increased 2.5 percent in the third quarter of 2026 compared with a year earlier, and output per worker increased 2.0 percent, its strongest growth since the first quarter of 2018 excluding the pandemic. The author adds that they are going to squeak through, and that immigration did not ruin the country and will not lead to civil war.
Broken World, from 404 Media, is a brief piece under a title announcing a theory of why everything is breaking.
The PSF end-of-year fundraiser, from Simon Willison's Weblog, reports in the author's capacity as a Python Software Foundation board director that the PSF has launched its end-of-year fundraiser. The PSF looks after the IP behind Python, operates PyPI, sponsors core developer sprints, supports community events including PyCon US, and employs staff. It also operates the Developers in Residence program, now with three security developers including the newly welcomed Brett Smith.
Sony's updated turntable is an easy entry into the world of vinyl, from TechCrunch, reports that Sony's updated turntable for beginners features automatic playback, USB connectivity, and aptX Adaptive Bluetooth.
Bull Shows Intel Xeon Diamond Rapids Package at SC26, from ServeTheHome, reports that Bull showed an Intel Xeon Diamond Rapids LGA9324 package in its Sequana XH31600 platform at Supercomputing 2026.
David Ellison says HBO will air critical Elon Musk documentary despite $1B investment in Skydance, from TechCrunch, reports that questions emerged over how Musk's sizable stake in the company might affect Skydance's plans to distribute a documentary highly critical of him.
Waymo Robotaxis Pulled From Freeways Indefinitely, from Slashdot quoting Ars Technica, reports that Waymo has temporarily suspended freeway operations across all its robotaxi markets after California regulators ordered the company off the state's highways over repeated incidents involving construction zones. The restriction follows a June recall of nearly 3,800 vehicles and subsequent software updates that failed to prevent further incidents, including a robotaxi driving past a California Highway Patrol rolling roadblock. Waymo told Ars that safety is at the foundation of everything it does, and that its ride-hailing service continues to operate on surface streets, serving over half a million trips each week.
Early Black Friday Deal: Claude Max and Grok Bot Codes ($200 Value) While Supplies Last, from Behind the Craft, promotes a month of free Claude Max and Grok Bot subscriptions for paid subscribers, with codes valid only for new customers of each tool. The post lists additional perks including $600 in codes for Granola, Lovable, Replit, Wispr Flow, Linear, and other tools, a personal AI course, and a Slack community. It disclaims that redeeming a code makes the paid membership non-refundable.
Sam Altman says his life is so weird that he's unphased by Andrew Garfield portraying him in upcoming movie, from TechCrunch, reports that Altman said it does not feel like it is in the top 10 weirdest things about his life.
Judge signs off on sale of Spirit Airlines' data to Google, from The Record, reports that Google can take possession of defunct Spirit Airlines' data for $10 million. The tech giant agreed to protections as it plans to use the data in AI projects.
The window closes on a set of intertwined themes: the pace of artificial intelligence and the prospect of recursive self-improvement, the software vulnerabilities and outages that arise as AI is pressed into increasingly old systems, and the governance questions that cut across both. The pieces converge on a repeated observation: institutions, infrastructure, and political economies are all moving more slowly than the models they are being asked to contain.
- Codeberg.org Not Reachable
- One swell foop
- Voyager 1 Reaches One Light-Day From Earth
- China dominates battery graphite, but this startup could break the monopoly
- Nvidia Earnings; Office Convergence; OS Wars, AI Edition
- Dreams of birds
- India’s Humyn Labs raises $47M to feed U.S. robotics companies the data they need
- Felicis leads $7.5M bet on Unsiloed AI to make enterprise documents usable by AI agents
- France Is Blaming Iran for the Fridge Cyberattack
- Was Achilles Conscious?
- Volunteer Dentists, the Revolutionary Guards and Rural Development on Iran’s Home Front
- Once More Unto The Breach - Another Year In Review
- This “One-in-a-Million” Fish Has a Rare Genetic Condition That Turns It Yellow
- Nvidia clarifies when it will — and won’t —keep financing neoclouds
- Quad9 Is Leaving France: Why DNS Resolvers Are the Wrong Place to Fight Piracy
- Cyberattack hits Norwegian energy giant Equinor, disrupting internal services
- Putting Pacing Into Practice
- Ice hockey governing body website defaced by hackers amid dispute over Russia's exclusion
- Data Contracts
- The PSF end-of-year fundraiser
- British hacker gets 6 year sentence for $11.6M crypto theft scheme
- How Cloudflare WAF helps protect against Java4Shell (CVE-2026-114938)
- British hackers face prison for GTA 6 data breach and crypto thefts
- Save the Parasites!
- How To Build The Future: Eric Glyman, Co-CEO of Ramp
- Russian cyber-espionage group used stolen cloud data to hack into organizations, Google says
- Quickly, There’s No Time
- Pre-auth RCE in Nginx HTTP/2 handling
- Deutsche Bank
- Broken World
- Quoting Daniel Stenberg
- Satya Nadella Says He's AGI-Pilled Now
- Why Can Some People Lower Their Cholesterol Through Dieting While Others Can’t?
- Sony’s updated turntable is an easy entry into the world of vinyl
- AMA: I'm Robert Warren, Head of Comms at bitcoin miner Elektron Energy
- Microsoft AI: Seven new world-class models at scalable price-performance
- OpenAI CEO Sam Altman says it’s ‘hard to imagine the AI industry without Jensen’
- NSA continues response to interruptions in classified internal network
- "Anthropic Is Not a Safe AI Company" – Coalition Challenges Anthropic Safety
- Wednesday assorted links
- Sam Altman warns against a future where AI, not humans, calls the shots
- Can Science Save the Ice Skaters?
- Jensen Huang's Fall Hits 'Open' Models In Boost to Anthropic and OpenAI
- Bull Shows Intel Xeon Diamond Rapids Package at SC26
- David Ellison says HBO will air critical Elon Musk documentary despite $1B investment in Skydance
- More banks says customers using AI agents are on their own — even if they get hacked
- Wait, is Google Antigravity good now?
- Waymo Robotaxis Pulled From Freeways Indefinitely
- Early Black Friday Deal: Claude Max and Grok Bot Codes ($200 Value) While Supplies Last
- Google’s AI feature in Search gets its own incognito mode
- The economics of investment expensing
- Pacing AI via Compute Reallocation
- Scottish HTTP Codes
- Genie 4 is now available
- UK productivity update
- Lawmakers call on White House to develop defensive AI cyber systems
- The Speech Patterns of Psychoses
- Cambridge Dictionary Names 'Human' Its 2026 Word of the Year
- Spend management platform Payhawk announces $280M raise, 2B valuation
- Snapchat now lets developers build AR Lenses with AI assistants like Claude Code and Cursor
- 'Java4Shell' Flaw Exposes OpenJDK to RCE via Base64
- Trump administration officials size-up a bad stretch for tech
- Homo Erectus Adopted a Jungle Lifestyle 200,000 Years Ago
- Sam Altman says his life is so weird that he’s unphased by Andrew Garfield portraying him in upcoming movie
- Judge signs off on sale of Spirit Airlines’ data to Google
- Bank CEOs, NSA Tell Congress That AI-Driven Patches Are Behind Outages
- Can “GTA V” Help Scientists Model Climate Change in L.A.?
- Palmer Luckey thinks China is ‘thrilled to have Jensen out,’ so he wants the ousted Nvidia CEO reinstated
- Palmer Luckey warns AI-powered cyberattacks are outpacing defenses: ‘It’s time to start running’

