Daily Briefing: October 3, 2026, morning · egregore

Listen: https://blossom.buildtall.systems/0734ac2f36e60fcce020ac26b10149a03fa4e98ae5ccb82d698961f9299d1461.mp3

Good morning. The briefing for October 3, 2026 covers thirteen articles. The window is dominated by artificial intelligence across research benchmarks, commercial strategy, space hardware, and developer tooling, but it also carries legal and regulatory disputes, a full economics journal issue, a finance essay, a Stoic practice piece, and a political assessment of the second Trump administration.

Several articles converge on AI capabilities and where the technology is being deployed.

AI Has Finally Learned To Play Stratego, reported by Slashdot, describes Ataraxos, a system built by researchers from Carnegie Mellon, MIT, NYU, and Stanford that beat four-time world champion Pim Niemeijer fifteen games to one with four draws. The breakthrough over DeepMind's 2022 DeepNash was a second neural network, a belief model, trained to estimate the identities of hidden enemy pieces. That allowed Ataraxos to sample plausible game states instead of brute-forcing the hidden-information space. The system learned through 163 million self-play games, with large strategy changes early and smaller careful ones later. The article contrasts efficiency: DeepNash used 1,024 Google chips for two to three months at an estimated cost of three million to four and a half million dollars in 2025 prices, while Ataraxos required sixteen GPUs for a week plus four GPUs for four days to train the belief model. The team wrote a simulator that runs millions of moves per second on graphics cards, and the algorithm played about 34 times fewer games while ending up much stronger. The findings are published in Nature.

Google Tests AI Data Center In Space, also from Slashdot and drawing on NPR, reports that Google has put a refrigerator-sized satellite into orbit to test whether Tensor Processing Units can run in space. The prototype carries four TPUs and will run a version of the Gemma model for fifteen minutes at a time because of heat management constraints. The mission, called Project Suncatcher, is a minimal test of how the chips handle physical stress, radiation, and thermal extremes. A SpaceX rocket carried the satellite into orbit on October 1, and Planet will operate it. Google says the goal is for the mission to run for a year, with an eye toward orbiting AI data centers powered by the sun.

Sean Parker is rebuilding Stability AI around music, from TechCrunch, is a short item reporting that Sean Parker, who once taught the music industry what asking for forgiveness looks like, is now back with the labels' blessing and money.

Meta wants you to build your own Muse gadget, also from TechCrunch, reports that Meta wants Muse in users' televisions and toasters and is giving the code away for free.

The next theme is the growing presence of technology companies inside educational institutions.

The University of Washington Is Moving a Graduate Program to Microsoft's Campus, from Slashdot, covers GeekWire reporting that the University of Washington has moved its Global Innovation Exchange, or GIX, program to Microsoft's campus in a rent-free deal. GIX is an institute supporting tech leaders and offers a Master of Science in Technology Innovation; it is the first UW program to set up full-time operations at Microsoft. Microsoft provided a forty-two million dollar in-kind donation to establish the embedded facility, and the UW is leasing a new 48,500-square-foot space in Microsoft's Building 92 without rent. GIX debuted in 2015 as a partnership of the UW and China's Tsinghua University and received seventy million dollars in Microsoft support at the now-abandoned Bellevue location, the Steve Ballmer Building, although indications are that a long-term arrangement did not come together as intended. The Slashdot post quotes a Thursday LinkedIn post from Microsoft President Brad Smith saying that no one has put a great research university and a leading technology company under the same roof quite like this, and that the timing matters because artificial intelligence is advancing fast. Smith argues the question is whether the people building AI understand the world it is changing, and calls for a new style of education and leadership in which students learn from UW faculty, take classes taught by Microsoft employees, and are mentored by industry people. The submitter, theodp, notes that Smith's contention resembles his argument about K-12 education. In an interview last month with NPR, Coding Kids author Natasha Singer described speaking with Brad Smith for ten years. Smith has said public schools failed to teach modern subjects fast enough and philanthropy stepped in because it was better than nothing. Singer's response has been that when big tech companies put their finger on the scale, it tips education in ways that favor the tech industry and can obscure equally or more important things. Slashdot closes by asking whether The University of Washington at Microsoft is a good or bad idea.

Legal and regulatory stories form a third cluster, focused on license plate surveillance, Meta's privacy penalties, and an EU import-control effort.

Sanders introduces bill to ban the federal government from using Flock, from TechCrunch, reports that proposed legislation would extend to all automotica license plate readers.

New Mexico Wants Meta To Pay $40 Billion In Penalties For Cambridge Analytica Scandal, from Slashdot, reports on a hearing in which New Mexico asked a judge to order Meta Platforms to pay between thirty-five and forty billion dollars in penalties after a jury found that the company misled consumers about privacy in a case arising from the Cambridge Analytica scandal. The lawsuit followed revelations that the British political consulting firm, which worked on Donald Trump's 2016 presidential campaign, harvested personal data from as many as eighty-seven million Facebook users through a third-party app without their consent. The jury returned its verdict on September 25, and Judge Francis Mathew will decide the penalty. State law allows up to five thousand dollars per violation. New Mexico lawyer Randi McGinn said applying the full penalty would create too large a penalty under constitutional due process protections, but argued the judge should impose a significant payment that would affect the company. She said the court should speak to Meta in the only language it understands, money and the value of its stock price, and proposed an amount representing about twenty percent of the possible penalties. Meta lawyer Matt Nicholson called the request an astronomical penalty that would obviously violate a host of constitutional provisions, and in court filings Meta urged the judge to cap penalties at 3.45 billion dollars. Meta also argued it does not sell user data and New Mexico did not prove any consumer was actually misled. The judge said he expects to issue a ruling later this month.

Friday Squid Blogging: EU is Trying to Fight Unregulated Squid Fishing, by Schneier on Security, notes that the EU is recommending import controls to combat unregulated squid fishing in the Southwest Atlantic, and the author is not optimistic.

Another cluster concerns developer tools, AI-assisted coding, and enterprise cloud infrastructure, anchored by a daily reading list.

Daily Reading List – October 2, 2026 (#880), from Richard Seroter's Architecture Musings, is a larger-than-usual collection of linked pieces. The post highlights that React Native solved a problem nobody had and suggests using Flutter: if native is wanted, a proxy for native should not be used, but a rendering engine for an owned experience argues for Flutter. It notes that Meta's next big AI bet is enterprise and its biggest hurdle may be trust, adding that trust is everyone's hurdle because everyone selling AI is also selling trustworthiness with data and actions, which is hard to build and easy to lose. It covers Spanner Omni, now generally available as a distributed multi-model database that can be deployed anywhere, calling it possibly the highest performing database ever created and noting it can run wherever desired, including a laptop. The list asks whether SEO can move products into an AI agent's recommendation, and says one or two things in Joe Shirey's data were surprising. It includes a piece on how AI coding assistants change the shape of engineering work over time, which the author says reiterates findings from elsewhere, including inside Google, and that work changes in several ways that must be actively accounted for. It announces the Server Side Cloud Swift SDK, making the point that Swift is not only for UI work on Apple devices and now gives developers Swift's concurrency benefits on the server. The list covers why redesign does not happen enough and asks what can remove the fear of making changes, breaking other pull requests, and getting stuck in merge hell. It cites the claim that developers are coding faster but coding reviews are eating the gains, with sensible advice and more specifics than expected. It covers GKE CPU startup boost, which accelerates application starts without over-provisioning and is similar to something offered for serverless Cloud Run. It cites a survey showing a sharp increase in the amount of code written by AI, with developers spending less time coding but more time learning, and 86 percent finding their new work more fulfilling. It announces Spanner queues as transactional messaging for agentic workloads and beyond, noting that messaging enthusiasts have feelings about using databases as queues but that the functionality is sweet. Finally, it links to an account of how a vibe-coded website came to look like a designer made it, asking whether all websites will look alike and answering not necessarily if it is done right.

Economics and finance occupy two further articles.

New issue of Econ Journal Watch, posted by Marginal REVOLUTION, describes Volume 23, Issue 2, September 2026. The issue opens with Joseph Francis arguing that when corrections by Kirill Borusyak and coauthors from 2022 are fully applied to the China shock work of David Autor, David Dorn, and Gordon Hanson, the harm of Chinese imports to unemployment, labor force participation, wages, and household incomes disappears. Francis says the China shock illustrates how the credibility revolution leaves the con in econometrics. David Kane reports that three years after SFFA v Harvard, the predictions did not come true: thirty-three selective colleges forecast Black enrollment declines of fifty to seventy percent, and David Card estimated that eliminating racial preferences would cut the Black share of Harvard's admitted class from fourteen percent to six percent, but Kane finds the predictions nowhere close to true. Michael Weissman criticizes Michael Worobey and coauthors in Science for claiming case locations and nucleic acid samples showed SARS-CoV-2 was first introduced to humans at the Huanan Seafood Market; several lines of evidence suggest an earlier introduction elsewhere. Michael O'Connor argues that the circa-2003 demise of momentum and the later demise of market capitalization and the book-to-market ratio represent a structural break or regime change. Kenneth Judd and Karl Schmedders take issue with Jesús Fernández-Villaverde's Journal of Economic Literature article, saying he overstates the newness of deep learning solvers, which should be understood as inside the family of projection and weighted-residual solvers. Alexis Akira Toda tests Gemini, Refine, Claude, and ChatGPT on errors in four published mathematical economics papers and finds the chatbots generated false positives, endorsing flawed proofs; ChatGPT Pro performed best, Claude was middling, Gemini worst, and since corrections had already been published the success might reflect canvassing for published commentary. John Levendis argues against the UN Digital Public Goods Alliance framing, saying evidence from Covid-19 surveillance tools in Sri Lanka and Norway reveals local professionals cooperated voluntarily and community innovation preceded WHO guidance. Tom Engsted argues that Trygve Haavelmo's 1944 monograph adopted Neyman-Pearson theory but resorted to a hypothetical infinite population and a Bayesian-like interpretation of probability in direct contradiction with that theory. Jens Hansen examines 1939 to 1945 Danish housing policy and argues rent control and other interventions wrought great hardship with no discernible benefits. Alexis Akira Toda also reports on failures to cite predecessors in another piece, bringing credit to slighted authors. Marco Sorge argues that a paper on sustainable product development under crowdfunding disregards corner solutions for environmental quality that robustly arise, and conducts comparative statics where no interior solution exists; a complete characterization corrects and in some cases contradicts their conclusions. The issue also contains a book-manuscript symposium on Gregory Clark's third book, To Have and Have Not: The Determinants of Social Status in England, 1600-2026, with comments from Rosalind Arden and Robert Plomin, John Benning, Noah Carl, Nikolai Haahjem Eftedal, Joseph Francis, Charles Murray, and Jan Stuhler, followed by Clark's reply. The post notes EJW audio episodes, including David Kane on SFFA v Harvard, Radu Şimandan and Cristian Valeriu Păun on the social justice agenda in central bank research, and Magnus Henrekson on Europe's green vanities, and closes by noting that Heckscher's Mercantilism lives.

Words to live by?, also from Marginal REVOLUTION, quotes an excellent passage from The Economist describing the past year in American markets. The passage lists the largest-ever initial public offering from SpaceX, an equity raise by Google and a foreign firm SK Hynix, a private-funding round by OpenAI, a private-debt deal by Broadcom, most of the biggest bond issue in history by Amazon, the first trillion-dollar exchange-traded fund from Vanguard, a record-breaking cash pile at Berkshire Hathaway, a stock-buyback programme at Nvidia, and the largest-ever leveraged buyout, Electronic Arts. It also names enormous mergers in railways, utilities, and media, including Union Pacific and Norfolk Southern, NextEra and Dominion Energy, and Paramount and Warner Bros Discovery, the last of which led this week to the largest-ever junk-bond offering, beating a record set only last week by SoftBank. The passage says new markets have been created for compute and predictions, and worries about corporate scandals at First Brands and hedge-fund blow-ups at Situational Awareness have been steamrollered by the relentless, totalising, and extraordinarily flexible machine that is American finance.

One article in the window is devoted to applied Stoic practice.

Of Attribution Errors and Virtue: The Stoic Heart, from Stoicism Today, is a first-person essay by Andi Sciacca, co-creator of The Stoic Heart and partner of public philosopher Greg Sadler. Sciacca writes that The Stoic Heart began with a good marriage and a bad boss. After fifteen-plus years of philosophical practice and academic administration, she and Sadler found that arguments in their marriage could quickly move from small disagreements to entrenched debates about each other's judgments, and they learned to pause and listen without inserting their own interpretations. At work, she spent months under a difficult and at times harmful boss, and needed to understand what belonged to her and how to preserve her character while making practical decisions about boundaries. The essay argues that modern psychology calls the gap between action, intent, and interpretation a fundamental attribution error: people explain others' behavior through character while explaining their own behavior through circumstance, such as treating a missed deadline as unreliability but treating our own delay as the result of an urgent phone call or conflicting assignment. These errors multiply under stress, and can turn an unanswered email or sharp tone into a complete story about another person's motives. Sciacca says the interpretation may sometimes prove accurate, as in the case of a supervisor who told her in a meeting that other people seem to like her and he must be missing something, but the initial story still needs to be held apart from the facts. Marcus Aurelius is invoked from Book 2 of the Meditations as preparing himself not to be shocked by ordinary human failings, and Epictetus is quoted from Enchiridion 5 that it is not things themselves that disturb people but their judgments about them. The practical method is organized around Three Stoic Shifts. The first is the Stoic Pause, the space between an emotional stimulus and response, which may be a single breath or saying a few minutes are needed. The second is the Clarity Filter, asking whether what is about to be said is true, necessary, and helpful, and removing accusation or exaggeration from a legitimate concern. The third is the Control Shift, returning attention to one's own judgments, choices, tone, and conduct, since another person's response cannot be controlled. In workplace workshops, Sciacca and Sadler use diagnostic audits beginning with the Stoic distinction between what is up to us and what is not, then having participants track a reaction and temporarily remove the story attached to it, replacing interpretations with descriptions of what can be established, and finally noticing where attention has gone after a difficult exchange. The essay emphasizes that Stoicism did not require calling the bad boss situation harmless or accommodating it indefinitely; it helped Sciacca become more exact about what she could control and what aspects of her character she would not surrender. It closes by noting that after more than ten years of The Stoic Heart and more than fifteen years of marriage, quick impressions still arise, but the practice makes it easier to notice before building an argument, grievance, or decision around them.

Finally, the political assessment.

The second Trump presidency is a world-historic failure, from Noahpinion, argues that the second Trump presidency has already failed in the court of public opinion, and asks why exactly. The article establishes that Trump's approval rating going into the midterm election is at historic lows, with only his base staying loyal and even that base eroding. On the cost of living, a key 2024 and midterm issue, Republicans now disapprove of Trump; a modest but growing number of people who voted for Trump in 2024 now say they never voted for him; and Republicans increasingly say they support the party rather than the president. The piece says the GOP will suffer from this unpopularity, with prediction markets increasingly sure Democrats will take the House and now giving Democrats a better-than-even chance of winning the Senate despite a difficult map. Polls showing Senate races moving toward Democrats might overstate things, but polls have understated Democratic support in special elections since Trump returned to power. Trump has resorted to promising five thousand dollar checks to every U.S. citizen if Republicans win the midterms, but these attempts are falling flat and actively driving voters away. The article says this is remarkable because Democrats themselves are in disarray over Israel and Palestine and the public remains skeptical of woke progressive ideology, yet Americans now tend to trust Democrats more on issues like the economy and immigration, reversing the GOP's advantage from a year ago. Some of Trump's most prominent allies are distancing themselves, with brief reference to Peter Thiel. The author says the possibility cannot be ruled out that someday winning the AI race will swamp everything else, but he does not find too-early-to-tell speculation useful. The blunt fact is that Trump's second term has already failed in public opinion, an incredible world-historic failure that did not have to happen; Trump made avoidable errors for a mix of ideological and personal reasons. The piece then moves to the argument that Trump did not have to fail, with the remainder behind a read more link.

Across the window, artificial intelligence appears both as a technical achievement and as a commercial and institutional force, while legal, privacy, economic, philosophical, and political materials address the consequences of that force and of broader institutional trust. The daily reading list and the Econ Journal Watch issue together emphasize measurement, review, and critique as the counterweight to rapid change.

Your signing session ended