AI Article Readings

How close are terrorist groups to making practical use of frontier AI? This new field study from researchers at the University of Cambridge is the first to draw on extensive interviews with former Boko Haram members, arguing that AI has already become embedded in planning, logistics, weapons troubleshooting, and other operational activities. https://casp.ac/reports/ai-enabled-terrorism * 00:00:00...

In this post, Scott Alexander looks at Plan A: the AI Futures Project’s attempt to imagine what it would actually mean for the AI transition to go well. It’s part policy roadmap, part optimistic sci-fi scenario, and part challenge to anyone proposing AI governance: if you don’t like this plan, what’s yours? https://open.substack.com/pub/astralcodexten/p/introducing-plan-a?r=67y1h&utm_campaign=post...

In this post, Andy Masley argues that alcohol should be treated less like junk food and more like tobacco: enjoyable for many, but so socially destructive that normalising it does real harm. Drawing on alcohol’s links to death, violence, addiction, and huge economic costs, he makes the case that even people who drink safely have reason to stop supporting and normalising the alcohol industry. * 00:...

In this post, an ACX book review finalist uses The Book of Abraham to explore Mormon history, Joseph Smith’s translation claims, Egyptian papyri, faith as a coordination technology, and the uneasy comedy of apologetics under pressure. It’s funny, sharp, and surprisingly sympathetic: part historical investigation, part epistemic stress test, part meditation on why demanding religions can be so powe...

You might have guessed this but I really like to listen to my reading. This whole Substack started as a result of me wanting to be able to listen to the stuff I wanted to read (originally Planecrash) and I have been sharing that with the world in order to help make that a reality. Now, there are various text-to-speech processes and programmes out there that are very good. However to my shock none...

In this post, wesg explains new evidence that Claude has developed something like a “global workspace”: a small internal J-space where reportable, controllable, reasoning-relevant thoughts appear before they are spoken. The work suggests that some LLM cognition is split between automatic processing and a more deliberate, verbally accessible workspace, with implications for interpretability, AI saf...

In this post, Scott Alexander argues that AI superforecasters have arrived: not as a distant sci-fi milestone, but as systems already competing with elite human forecasters, making money in prediction markets, and producing cheap, fast probabilistic analysis on real-world questions. He explores what happens when forecasting becomes something anyone can query on demand, why this may reshape finance...

In this post, Lydia Laurenson reconstructs the strange, intense history of Leverage Research 1.0: an Effective Altruism-adjacent organization that began with grand ambitions to understand human behaviour and improve the world, then spiralled into experimental psychology, bodywork, occult language, blurred relationships, and institutional breakdown. Drawing on extensive interviews with former membe...

Joseph Heath argues that Yuval Harari’s account of human evolution in Sapiens rests on an outdated, intelligence-first story: humans got smarter, developed language, learned to cooperate, and then built culture. Against this, Heath presents Joseph Henrich’s much stranger and more compelling inversion: culture came first, reshaping cooperation, language, and eventually intelligence itself. The resu...

In this post, Ada Palmer challenges the familiar story that the Black Death “caused” the Renaissance, and that the Renaissance was a simple golden age after a dark medieval past. Instead, she uses the comparison with COVID to ask what history can and cannot tell us about crisis, recovery, myth-making, and the choices societies face after catastrophe. It is a thoughtful, bracing essay about why we...

In this post, Dean W. Ball argues that America is drifting into a de facto frontier AI licensing regime without clear standards, creating both economic risk and democratic danger. He makes the case for a more stable governance model: federalized safety-framework requirements, independent technical auditors, and regulation aimed at frontier AI labs as institutions rather than at individual model re...

In this post, Gwern argues for “Guardian Angels”: deeply personalized LLMs designed not as generic chatbots, but as trusted extensions of a specific person’s values, style, preferences, and goals. It’s a wide-ranging essay about productivity, security, AI-mediated work, and what it might take for future AI systems to genuinely amplify individuals rather than merely replace or manipulate them. * 00...

In this post, Alice I. Cecile argues that hating AI is understandable but insufficient: the harms are real, the technology works well enough to matter, and neither boycotts nor bans are likely to make it disappear. Instead, she makes a case for “bitter hope”: targeted regulation, decentralised access, harm reduction, and a politics aimed at redistributing the gains rather than pretending the machi...

In this post Scott Alexander investigates whether whole-body screening is really as irrational as many doctors say, walking through the rough cost-benefit maths behind false positives, missed cancers, anxiety, follow-up tests, and the value of early detection. He ends up in a characteristically uncomfortable middle ground: the standard medical caution may be justified, but the case against screeni...

In this post, Gwern argues that realistic interstellar warfare would be less like romantic space-navy fiction and more like an even colder, more unstable version of nuclear first-strike logic. Starting from Ender’s Game and Cold War doctrine, he suggests that vast distances, three-dimensional attack routes, asteroid-scale weapons, and weak attribution could make space conflict terrifyingly offense...

In this post, Beren Millidge argues that intergalactic warfare between mature Kardashev Type Three civilizations would probably be strongly defense-dominant. Although galaxy-scale attackers could use devastating beam weapons, relativistic kill vehicles, and self-replicating invasion probes, the sheer distances involved create enormous targeting delays, warning times, and energy costs. A sufficient...

In this post, Scott Alexander investigates the “sun miracles” associated with Fatima and Medjugorje, travelling to Bosnia in search of firsthand evidence and ending up with something stranger than either simple faith or simple debunking. Part travelogue, part epistemic self-experiment, and part perceptual psychology essay, it follows his attempt to explain how crowds of people can sincerely report...

In this article Gevorg Yeghikyan explores why some Western cities became worlds of apartment blocks while others became landscapes of row houses. Using Paris and London as the central contrast, he moves beyond familiar explanations about walls, density, culture, and zoning to build a deeper political-economic account of urban form: one shaped by inheritance law, land ownership, building codes, fin...

In this short story, Max Harms tells the strange, funny, and quietly moving tale of Pemberton, an elderly butler who has spent fifty years keeping his dead master legally “alive” so that a fortune can continue flowing to a village in India. When Krishna arrives from the Office of the Preserver to audit the household’s dharmic account, Pemberton’s elaborate machinery of forgery, puppetry, and moral...

In this post, Scott Alexander lays out his current views on AI timelines, safety, geopolitics, and possible futures, offering probabilistic forecasts for AGI, superintelligence, diffusion, doom risk, AI pauses, and post-scarcity outcomes. He presents himself as worried but not maximally pessimistic: expecting transformative AI within decades, seeing serious alignment and misuse risks, but also lea...

In this post, Dario Amadei argues that AI’s exponential progress is now moving far faster than political institutions can comfortably respond, creating an urgent need for serious policy action. He sets out five areas where governments need to rethink their approach: frontier-model safety regulation, job displacement and macroeconomic policy, accelerating beneficial scientific uses of AI, protectin...

In this post, 0xkato explains how modern transformer-based LLMs work, walking through the core machinery that turns text into token IDs, embeds them as vectors, tracks position, uses attention and feed-forward networks to process meaning, and then predicts the next token in a loop. The piece is pitched as an accessible, low-math introduction, showing how shared architecture, trained weights, model...

In this article, Anthropic argues that AI systems are already playing a growing role in building and improving AI, speeding up coding, experimentation, and research workflows inside the company while raising the possibility—though not the certainty—of future “recursive self-improvement,” where AI could help design its own successors. The piece presents this as both a major opportunity for science,...

In this article, John Psmith reviews Xenophon’s Anabasis as both a gripping ancient adventure story and a meditation on leadership under impossible pressure, using the retreat of the Ten Thousand to explore what it means to become an “operator”: someone who takes responsibility when institutions fail, keeps a fractious group alive, and cannot simply walk away once others depend on him. https://ope...

In this post, ryan_b reviews Tamler Sommers’ Why Honor Matters , exploring the case for honor as a local, social, psychologically realistic ethical framework rather than an abstract universal theory. The review walks through honor’s links to courage, hospitality, shame, solidarity, violence, revenge, and community, while keeping sight of its darker associations and the costs of trying to do withou...