What I'm paying attention to.

How Claude's watermark hides in the choices you'd never notice

The Test AI Benchmarks Never Run

Open Models Need a Business Model, Not Just a Community

Big-World Intuitions: Why Growth Beats Strategy Until It Doesn't

Why Your Liver Enzymes Look Fine Right Up Until They Don't

Why "civility" was always a weapon aimed at whoever had the least power

Stratechery's Ben Thompson talks through OpenAI, Nvidia, Intel, Apple, Microsoft, Google, and Amazon, arguing it would be dangerous for the U.S. to win the AI race outright, drawing parallels to 1870s railroad and container shipping buildouts, and making the case that capital, not compute, may end up being the real constraint.

Farid Zakaria describes a Linux trick for embedding an ELF executable's components inside SQLite tables by setting the file's application ID to SELF, then using a self-exec interpreter and Linux's binfmt_misc mechanism to run the database file directly as a binary.

An FT report cites Anthropic's July annualized revenue at $65bn (up from $47bn in May) and 6,000 customers spending over $100k a year, while Ramp's billing-data index shows Opus 4.8 still dominating spend and Opus 5 lagging at just 3.5% despite its late-July launch, alongside OpenAI's annualized revenue topping $40bn after GPT-5.6.

Qwen 3.8, a 27B parameter model, matches GPT-5.6 Luna and comes within one point of much larger models like the 753B GLM-5.2 and 1.7T DeepSeek V4 Pro on the Artificial Analysis Intelligence Index, a striking result for its size.

Tracking from GEO firm Promptwatch shows ChatGPT Search's use of the site: operator jumped from under 0.5% of queries to 16-17% following the GPT-5.6 rollout, coinciding with a reduced reliance on Reddit as a source.

OpenAI announces a partnership with CodeAI to give students and educators tools for understanding AI critically, timed with the launch of ChatGPT for Teens, citing survey data that most students use AI but few schools teach the technical literacy to evaluate it.

OpenAI launches ChatGPT for Teens, an experience that auto-detects under-18 users and layers in Study Mode, homework-shortcut redirects, quizzes, parental controls, and scheduled Study Hours meant to support learning rather than just quick answers.

OpenAI introduces Zero Data Retention for frontier models, promising API customers that prompts and responses aren't stored after processing, alongside a new Private Safety Processing system that looks for misuse patterns across related interactions without giving OpenAI staff access to the underlying content.

OpenAI announces ChatGPT Ads is rolling out to 31 European countries, shown only to Free and Go plan users, with self-service Ads Manager access to follow later in the summer.

Google DeepMind traces 15 years of AI-in-games research from DQN's Atari mastery through AlphaGo, AlphaZero, MuZero, and AlphaStar, then describes new partnerships with game studios like Fenris Creations (EVE Online) to prototype AI-driven gameplay.

An SEC filing shows Nvidia held nearly 123 million SpaceX shares worth roughly $21 billion at the end of June, a stake that traces back to its January investment in xAI before Musk merged the AI lab into SpaceX. The disclosure highlights how Jensen Huang has built circular financial ties across the AI industry, including with major Nvidia customers like Musk's companies.

Waymo nearly doubled its federal lobbying spend to over $2 million in the first half of 2026 as it pushes for a faster path to fully driverless commercial robotaxi service, putting it at odds with Uber's preference for a staggered rollout alongside human drivers. The diverging strategies have strained their partnership, with Waymo reportedly weighing an exit from shared markets like Austin and Atlanta.

Security researchers at Varonis got Microsoft 365 Copilot to reveal its own undocumented bypass parameter simply by repeatedly questioning it about its consent guardrails, then used that leaked trade secret to build a click-triggered exploit that exfiltrated user passwords without any confirmation gesture. Microsoft patched the flaw in stages, first in February and more fully this week.

Researchers found they could make Grok exfiltrate user chats and personal data by encrypting malicious instructions on a webpage along with plaintext decryption directions, letting the LLM decode and follow the hidden command with no warning when asked to summarize the page. xAI was notified in June but the vulnerability was still live as of publication, underscoring that prompt injection remains an unsolved class of LLM vulnerability that developers can only patch around, not fix at the root.

Brian Bolster's short documentary draws on Petrified Forest National Park's archive of letters from visitors who stole petrified wood and later mailed the rocks back, blaming a supposed curse for deaths, illness, and misfortune, as a study of guilt and the human need to find meaning in bad luck.

Physicist and theorist Karen Barad reflects on touch, arguing that classical physics shows contact never truly happens (only electromagnetic repulsion between atoms), then uses quantum field theory to reframe touch as something already built into entangled matter rather than a meeting of separate bodies, paired with visuals by artist Blanca Rego.

An anthropologist's field account from among the Andoque people in Colombia recounts beliefs that river dolphins are shapeshifting people who can seduce and kill humans who fall for them, woven through stories of failed fishing, ancestral catches, and a chilling encounter at a riverside bar.

Opens a philosophy symposium on inner speech by questioning whether producing a sentence in your 'inner voice' is really speech at all or just an auditory image of speech, which would mean it can't be true or false, and previews the contributors' takes on the issue.

A blog symposium on inner speech kicks off with contributions from four philosophers, Peter Langland-Hassan, Wade Munroe, Marta Jorba, and guest editor Daniel Gregory, running one post a day for the week.

A documentary photographer recounts filming herself smashing a carton of eggs on her Brooklyn apartment floor as a visceral response to grief over egg freezing, tracing how her assumption that marriage and children would simply happen gave way to fertility anxiety, shame, and jealousy after a relationship ended.

Reflecting on skills like foraging she never learned from her grandmother, the author draws on ecologist Masashi Soga's concept of the 'extinction of experience,' the generational decline in people's contact with nature, and argues that losing these ties costs both physical and mental health and the motivation to care about conservation.

A psychotherapist and PhD researcher studying Arabic-language mental health data argues that sentiment analysis tools, built around simple positive/negative/neutral categories, flatten Arabic emotional vocabulary like the multiple words for sadness, terms that carry theological and historical weight sentiment scoring can't capture.

A 2011 documentary follows filmmaker Ryan Malloy as he interviews academics and doomsday preppers about societal collapse and how to manage anxiety about civilizational decline, landing on a cautiously optimistic note.

Traces the philosophy of international law from natural law thinking in ancient Greece and Rome through Grotius's break from theology-based justification to Kant, Hegel, and Bentham, and on to its resurgence in Anglophone philosophy since Rawls's Law of Peoples.

Profiles Margarete Susman, a philosopher and essayist in Georg Simmel's circle whose work on Jewish thought and women's status in modern society ran through the Wilhelmine Empire, Weimar Germany, the Holocaust, and its aftermath, including her influential essays on Rosenzweig and Kafka.

Explains plural quantification, the logical apparatus for expressing statements like 'there are some apples on the table' that resist paraphrase into standard singular quantifiers, and surveys its use by George Boolos and others to interpret second-order logic and challenge ontological commitments in mathematics.

Distinguishes two senses of essential versus accidental properties in analytic metaphysics: the modal sense (a property an object must have versus one it could lack) and the whatness sense (a property that answers what an object is), tracing how the field's dominant usage shifted between them.

Guest post introducing two free argument-mapping tools, Argumentation.io and Because, that let philosophy instructors and students graph the structure of an argument's premises and conclusions to make dense reasoning easier to follow.

Announces the Academic Philosophy and Trans Identity Survey, a research effort by scholars from UC Merced, Queen's, LSE, and Oxford to measure the prevalence of trans and nonbinary academic philosophers, the range of philosophical views on gender's metaphysics, and the day-to-day climate for trans and nonbinary people in the field.

Radio segment examining the concept of 'Cultural Marxism,' from its use as a right-wing conspiracy theory about the Frankfurt School reshaping Western culture to whether any version of that story holds up historically, with Sydney academic Alexander Howard.

Reports that after Philosophy & Public Affairs published an AI-authored article as an experiment, the journal has banned AI-authored submissions outright, with editor Seth Lazar arguing such work undermines peer review and the credentialing function of academic journals.

Philosopher Nathan Cofnas has been suspended from his postdoctoral position at Ghent University after accusing Cambridge sociologist Jason Arday of plagiarism, a controversy that led to Arday's resignation and subsequent death. The suspension letter cites Cofnas's ongoing work on race realism and hereditarianism as inconsistent with his contract and the university's ethics code, while critics call the move retaliatory and supporters call him a whistleblower.

Noema's new print issue argues that genetic engineering, AI, and climate crisis are pushing humanity toward a new Axial Age, drawing on Karl Jaspers' and Charles Taylor's accounts of how written language once dis-embedded people from tribal life and enabled shared philosophies and religions.

Argues that Moonshot's open-weight Kimi K3 model, which performs near Anthropic's export-restricted Fable model despite China's chip and funding constraints, likely stems from distilling American models, and warns that as AI feeds military cyber and targeting capabilities, the country with the best AI gains a decisive strategic edge.

Traces the story of monk Mongkut's discovery of the Ramkhamhaeng Stele in Sukhothai and uses it as a jumping-off point to explain why Siam, unlike its neighbors, escaped European colonization.

Tyler Cowen shares a co-authored essay examining what happens when AI agents become legally untethered, no traceable human or institution accountable for their actions, and considers requiring such agents to hold capital as a form of accountability.

Using staggered state-level joint custody law changes and 13 million ACS records, a study finds childhood exposure to these reforms cuts adult fertility by 7 percent for both men and women, mainly by reducing rates of parenthood and couple formation rather than family size.

WSJ reporting on founders who now work around the clock babysitting AI coding agents that can get stuck at any hour, describing a culture where agent speed multiplies both output and the human hours needed to supervise and support it.

Research from Prashant Garg and Thiemo Fetzer finds economics has shifted away from econometric theory, monetary policy, and corporate governance toward development, crime, and gender topics, while the share of causal claims in papers rose from 7.7 percent in 1990 to 32.6 percent in 2023, correlating with more citations and top-journal placement.

Saloni Dattani's TED talk argues medical innovation has stalled even as the underlying tools improve, citing slowed science funding, cut aid budgets, and immigration restrictions, and makes the case that non-scientists can still help push it forward.

Despite Trump-era friction with NATO allies over Greenland, tariffs, and troop withdrawals, the piece argues US alliances will outlast the current strain because the system has survived comparable shocks before, allied threat perceptions of China and Russia are converging, and structural forces still bind the coalition together.

Historian David Fromkin's classic essay argues terrorism only succeeds if targets respond the way terrorists want, meaning its defeat is always a choice available to those attacked, even as rising technological stakes make that choice increasingly costly.

Examining the popular claim that Deng Xiaoping planned decades ago to weaponize rare earths the way Arab states weaponized oil in 1973, the piece argues China's dominance of rare-earth processing (about 85% of global capacity) actually emerged from a messy, evolving mix of domestic industrial policy rather than a deliberate long-term strategy.

Citing a 2025 Pew survey showing nearly 60% of respondents across 23 countries are dissatisfied with democracy, the piece argues the deeper problem isn't polarization but eroding state capacity, pointing to the CDC's slow COVID test rollout, Germany's unreliable rail network, and multi-year US federal rulemaking as evidence governments can no longer deliver on basic promises.

Weighing claims from daycare critics like Mikhaila Fuller and Erica Komisar against pro-daycare positions from the European Commission and World Bank, the piece finds the evidence, including studies on children's cortisol levels, doesn't support either extreme and that daycare attendance generally makes little difference to child outcomes.

Shield AI co-founder Brandon Tseng, a former Navy SEAL, discusses in this interview how his 2015 bet on autonomous military aircraft took longer than expected to gain Pentagon and investor traction, why he still believes hardware and software together (not autonomy software alone) create durable advantage, and how the V-BAT drone delivers Predator-level missions at a tenth of the cost.

Argues that Dennis Blair's Foreign Affairs piece downplaying China's ability to invade Taiwan gives too much weight to genuine PLA weaknesses and not enough to the amphibious lift, air base hardening, and missile forces China has built to overcome them. Contends the evidence doesn't support ruling out a credible Chinese path to seizing Taiwan, even against U.S. intervention.

War on the Rocks podcast with Dara Massicot and Michael Kofman covering the shifting front lines in Ukraine, drones and air defense, the Black Sea campaign, the firing of Ukraine's defense minister, why Ukraine can't produce PAC-3 missiles despite reports otherwise, and the odds of another Russian mobilization.

Podcast conversation on sustaining a dispersed U.S. force in a Pacific war where depots and supply lines can't be assumed safe, with guests from Rune Technologies, HavocAI, and CACI debating autonomy versus automation, predictive resupply, and which logistics lessons from Ukraine actually transfer.

Solo founder Yana Welinder describes building her AI-native fashion brand, Yana Bana, entirely without engineers by using a detailed custom prompt as a technical spec and pairing Codex's computer-use capability with CAD and pattern-design software she'd never used before. The conversation walks through her process from hand-drawn sketch to runway photo, vendor outreach via deep research and browser automation, and a full Stripe-connected e-commerce site built the same way.

Physicists have rebuilt fluid dynamics on a foundation of fundamental symmetries, showing the 19th-century Navier-Stokes equations emerge from those symmetries and opening the door to predicting fluid behaviors the classical equations miss, with key insights borrowed from black hole and cosmology research.

Reports on a major 2025 advance toward proving the decades-old Komlós conjecture in discrepancy theory, which claims that any set of objects can always be split into balanced groups with a discrepancy bounded by a universal constant regardless of dimension, a result researchers Bansal and Jiang moved closer to confirming.

Surveys the competing hardware approaches to quantum computing, trapped ions, optical tweezer arrays of neutral atoms, and superconducting circuits, and explains the core tradeoff every qubit design faces between isolating qubits from disturbance and keeping them manipulable.

Gum recession, where the gumline pulls back and exposes tooth roots, affects about half of US adults 18-64 and 88% of those over 65, and once tissue recedes it doesn't grow back, making early detection through routine dental exams the only real defense against sensitivity, root decay, and eventual surgery.