Understandary

•

Canale • 2 podcast

Notizie: approfondimenti

Notizie: approfondimenti

Ogni settimana
Cultura e società

Cultura e società

Ogni 2 settimane

23/12/2025

Data Center Politics

This week we talk about energy consumption, pollution, and bipartisan issues. We also discuss local politics, data center costs, and the Magnificent 7 tech companies. Recommended Book: Against the Machine by Paul Kingsnorth Transcript In 2024, the International Energy Agency estimated that data centers consumed about 1.5% of all electricity generated, globally, that year. It went on to project that energy consumption by data centers could double by 2030, though other estimates are higher, due to the ballooning of investment in AI-focused data centers by some of the world’s largest tech companies. There are all sorts of data centers that serve all kinds of purposes, and they’ve been around since the mid-20th century, since the development of general purposes digital computers, like the 1945 Electronic Numerical Integrator and Computer, or ENIAC, which was programmable and reprogrammable, and used to study, among other things, the feasibility of thermonuclear weapons. ENIAC was built on the campus of the University of Pennsylvania and cost just shy of $500,000, which in today’s money would be around $7 million. It was able to do calculators about a thousand times faster than other, electro-mechanical calculators that were available at the time, and was thus considered to be a pretty big deal, making some types of calculation that were previously not feasible, not only feasible, but casually accomplishable. This general model of building big-old computers at a center location was the way of things, on a practical level, until the dawn of personal computers in the 1980s. The mainframe-terminal setup that dominated until then necessitated that the huge, cumbersome computing hardware was all located in a big room somewhere, and then the terminal devices were points of access that allowed people to tap into those centralized resources. Microcomputers of the sort of a person might have in their home changed that dynamic, but the dawn of the internet reintroduced something similar, allowing folks to have a computer at home or at their desk, which has its own resources, but to then tap into other microcomputers, and to still other larger, more powerful computers across internet connections. Going on the web and visiting a website is basically just that: connecting to another computer somewhere, that distant device storing the website data on its hard drive and sending the results to your probably less-powerful device, at home or work. In the late-90s and early 2000s, this dynamic evolved still further, those far-off machines doing more and more heavy-lifting to create more and more sophisticated online experiences. This manifested as websites that were malleable and editable by the end-user—part of the so-called Web 2.0 experience, which allowed for comments and chat rooms and the uploading of images to those sites, based at those far off machines—and then as streaming video and music, and proto-versions of social networks became a thing, these channels connecting personal devices to more powerful, far-off devices needed more bandwidth, because more and more work was being done by those powerful, centrally located computers, so that the results could be distributed via the internet to all those personal computers and, increasingly, other devices like phones and tablets. Modern data centers do a lot of the same work as those earlier iterations, though increasingly they do a whole lot more heavy-lifting labor, as well. They’ve got hardware capable of, for instance, playing the most high-end video games at the highest settings, and then sending, frame by frame, the output of said video games to a weaker device, someone’s phone or comparably low-end computer, at home, allowing the user of those weaker devices to play those games, their keyboard or controller inputs sent to the data center fast enough that they can control what’s happening and see the result on their own screen in less than the blink of an eye. This is also what allows folks to store backups on cloud servers, big hard drives located in such facilities, and it’s what allows the current AI boom to function—all the expensive computers and their high-end chips located at enormous data centers with sophisticated cooling systems and high-throughput cables that allow folks around the world to tap into their AI models, interact with them, have them do heavy-lifting for them, and then those computers at these data centers send all that information back out into the world, to their devices, even if those devices are underpowered and could never do that same kind of work on their own. What I’d like to talk about today are data centers, the enormous boom in their construction, and how these things are becoming a surprise hot button political issue pretty much everywhere. — As of early 2024, the US was host to nearly 5,400 data centers sprawled across the country. That’s more than any other nation, and that number is growing quickly as those aforementioned enormous tech companies, including the Magnificent 7 tech companies, Nvidia, Apple, Alphabet, Microsoft, Amazon, Meta, and Tesla, which have a combined market cap of about $21.7 trillion as of mid-December 2025, which is about two-thirds of the US’s total GDP for the year, and which is more than the European Union’s total GDP, which weighs in at around $19.4 trillion, as of October 2025—as they splurge on more and more of them. These aren’t the only companies building data centers at breakneck speed—there are quite a few competitors in China doing the same, for instance—but they’re putting up the lion’s share of resources for this sort of infrastructure right now, in part because they anticipate a whole lot of near-future demand for AI services, and those services require just a silly amount of processing power, which itself requires a silly amount of monetary investment and electricity, but also because, first, there aren’t a lot of moats, meaning protective, defensive assets in this industry, as is evidenced by their continual leapfrogging of each other, and the notion that a lot of what they’re doing, today, will probably become commodity services in not too long, rather than high-end services people and businesses will be inclined to pay big money for, and second, because there’s a suspicion, held by many in this industry, that there’s an AI shake-out coming, a bubble pop or bare-minimum a release of air from that bubble, which will probably kill off a huge chunk of the industry, leaving just the largest, too-big-to-fail players still intact, who can then gobble up the rest of the dying industry at a discount. Those who have the infrastructure, who have invested the huge sums of money to build these data centers, basically, will be in a prime position to survive that extinction-level event, in other words. So they’re all scrambling to erect these things as quickly as possible, lest they be left behind. That construction, though, is easier said than done. The highest-end chips account for around 70-80% of a modern data center’s cost, as these GPUs, graphical processing units that are optimized for AI purposes, like Nvidia’s Blackwell chips, can cost tens of thousands of dollars apiece, and millions of dollars per rack. There are a lot of racks of such chips in these data centers, and the total cost of a large-scale AI-optimized data center is often somewhere between $35 and $60 billion. A recent estimate by McKinsey suggests that by 2030, data center investment will need to be around $6.7 trillion a year just to keep up the pace and meet demand for compute power. That’s demand from these tech companies, I should say—there’s a big debate about where there’s sufficient demand from consumers of AI products, and whether these tech companies are trying to create such demand from whole cloth, to justify heightened valuations, and thus to continue goosing their market caps, which in turn enriches those at the top of these companies. That said, it’s a fair bet that for at least a few more years this influx in investment will continue, and that means pumping out more of these data centers. But building these sorts of facilities isn’t just expensive, it’s also regulatorily complex. There are smaller facilities, akin to ENIAC’s campus location, back in the day, but a lot of them—because of the economies of scale inherent in building a lot of this stuff all at once, all in the same place—are enormous, a single data center facility covering thousands of acres and consuming a whole lot of power to keep all of those computers with their high-end chips running 24/7. Previous data centers from the pre-AI era tended to consume in the neighborhood of 30MW of energy, but the baseline now is closer to 200MW. The largest contemporary data centers consume 1GW of electricity, which is about the size of a small city’s power grid—that’s a city of maybe 500,000-750,000 people, though of course climate, industry, and other variables determine the exact energy requirements of a city—and they’re expected to just get larger and more resource-intensive from here. This has resulted in panic and pullbacks in some areas. In Dublin, for instance, the government has stopped issuing new grid connections for data centers until 2028, as it’s estimated that data centers will account for 28% of Ireland’s power use by 2031, already. Some of these big tech companies have read the writing on the wall, and are either making deals to reactivate aging power plants—nuclear, gas, coal, whatever they can get—or are saying they’ll build new ones to offset the impact on the local power grid. And that impact can be significant. In addition to the health and pollution issues caused by some of the sites—in Memphis, for instance, where Elon Musk’s company, xAI, built a huge data center to help power his AI chatbot, Grok, the company is operating 35 unpermitted gas turbi

23/12/2025

•

17 min
13 GEN

Operation Absolute Resolve

This week we talk about Venezuela, Maduro, and international law. We also discuss sour crude, extrajudicial killings, and Greenland. Recommended Book: The Keep by F. Paul Wilson Transcript Back in mid-November of 2025, I did an episode on extrajudicial killings, focusing on the targeting of speedboats, mostly from Venezuela headed toward the United States, by the US military. These boats were allegedly carrying drugs meant for the US market, and the US government justified these strikes by saying, basically, we have a right to protect ourselves, protect our citizens from the harm caused by these illegal substances, and if we have to keep taking out these boats and killing these people to do that, we will. There’s been a lot of back-and-forthing about the legitimacy of this approach, both in the sense that not all of these boats have been shown to be carrying drugs, some just seemed to be fishing boats in the wrong place at the wrong time, and in the sense that launching strikes without the go-ahead of Congress in the US is a legally dubious business. There was also the matter of some alleged follow-up strikes, which seemed to be intended to kill people who survived the initial taking-out of the boats, which is a big international human rights no no, to the point of potentially being a war crime. All of this happened within the context of a war of words between US President Trump’s second administration and the increasingly authoritarian regime of Venezuelan President Nicolás Maduro, who followed the previous president Hugo Chávez as his hand-picked successor, and has more or less completed the authoritarian process of dissolving, coopting, or diminishing all aspects of the Venezuelan government that might ever check his power, which allowed him, in 2024, to bar the very popular, now Nobel Peace Prize winning candidate María Corina Machado from running, and her sub-in candidate, like previous Maduro opponent Juan Guaido, seems to have won the election by a fair bit, and in an internationally provable way, but Maduro’s government faked results that made it look like he won, and his single-party rule has since continued unabated. Or rather, it continued unabated until the early morning of January 3, 2026, around 2am, when US Operation Absolute Resolve kicked into action, leading to the—depending on who you ask—justified captured or illegal kidnapping—of Maduro and his wife from a stronghold in his country. And that’s what I’d like to talk about today: the operation itself, but also the consequences and potential meaning of it within the context of other important things happening in the world right now. — Maduro is immensely popular with about a fifth of the Venezuelan population, but essentially everyone else is strongly opposes him and his iron-fisted rule. It’s estimated that between 2017 and 2025, just shy of 8 million people, which is more than 20% of Venezuela’s 2017 population, has fled the country in order to escape a tyrannical government and its failed policies, which have collapsed the economy, made getting working and feeding oneself and one’s family difficult, and made crime, conflict, and the state-sanctioned oppression of anyone who doesn’t kowtow to the ruling party a commonplace thing. Trump speculated about the possibility of invading Venezuela even in his first administration, and part of the overt rationale was that it’s run by a failed government that most of the locals hate, so it would be an easy win. That justification shifted to orient around immigration and drugs by his second administration, and then more recently, Trump has said publicly that the real issue here is that Venezuela stole a bunch of US company-owned oil assets when it nationalized the industry back in the day, and those assets should be recaptured, given back to the US. Operation Absolute Resolve took months to plan and only about two and a half hours to complete. By most objective measures it was a spectacular military and intelligence success, especially considering all the moving parts and thus, all the things that could have gone wrong. The operation apparently involved at least 150 aircraft of various sorts, a spy within Maduro’s government, and months of surveillance, which helped them establish Maduro’s habits and routines, and that allowed them to map out where he would be, when, and what to expect going in to get him. All of these patterns changed in September of 2025 when US warships started massing in Caribbean, as Maduro started to get a little paranoid—justifiably, as it turns out—and he started moving between eight different locations, seldom sleeping in the same place more than one night in a row. He was eventually grabbed from a military base in Caracas, Venezuela’s capitol, and to make that happen the US military assets in the area had to take out local aviation and air defenses so that US Delta Force troops could be carried in by helicopter. Several air bases and communications centers were taken out by missiles, and fighter jets were bombed on air base tarmacs. Trump alluded that a cyberattack of some kind might have also been used to take out power in the area, though satellite imagery suggests bombs might have been used against a power station to make that happen. The operation apparently went almost exactly as planned, though a helicopter was damaged and the Delta Force team killed a large part of Maduro’s security team when he refused to surrender. A few US soldiers were wounded, but none were killed, and Venezuelan officials said, in the aftermath, that lat least 40 Venezuelans were killed throughout the country during the operation. Maduro and his wife were swept from the base before they could lock themselves in their safe room, and they were tucked into the helicopters which headed out to sea, landing them on the USS Iwo Jima, which is an assault ship. All of this took a matter of hours and, again, is generally considered to be an objective success, in terms of precision, outcome, and other such metrics. Morally, legally, and politically, however, the operation is receiving a far more mixed response, and that response is continuing to play out as Maduro works his way through a bizarre version of the US justice system where he’s being sent to court for drug dealing. In the US, Trump supporters have generally said all of this was a good, smart move, though some maintain that US involvement in any kind of international conflict is a waste of time, effort, and resources, and they worry about getting bogged down in another Iraq or Afghanistan-style conflict. Everyone else is generally against the effort, even those who admit that Maduro was a tyrant who needed to go—it’s good that he’s gone, but the way in which it was done is not just questionable, but worrying because of what it says about Trump’s capacity to unilaterally launch kidnapping missions against the leaders of other countries. Not a good look, but also kind of scary. Internationally the response is generally aligned with the latter opinion, especially in other countries that Trump has at some point threatened, which is most of them. Governments in South and Central America have been especially concerned, however, because one of Trump’s newer messaging efforts has revolved around the concept of a Western Hemisphere basically owned and protected by the US. Do whatever you want in the rest of the world, basically, but everything over here is ours. This has raised the possibility that an emboldened Trump might attempt similar maneuvers soon, including possibly claiming the Panama Canal for the US again, or grabbing the leaders of other Latin American countries he doesn’t think are kowtowing enthusiastically enough; toeing the new international line that he’s drawing, basically. He’s also renewed messaging around the possible purchase or capture of Greenland, which has been raising alarm bells across Europe in particular. Greenland is considered to be a vital strategic base for US security, and it would grant potential access to an abundance of also strategically and economically important minerals, both on land and underwater, but Greenland is an autonomous territory of Denmark, and most European leaders have said something along the lines of “if the US takes action to militarily claim Greenland, that’ll be the end of NATO,” an organization that was originally founded to help protect the world, and Europe especially, from military conquest from the Soviet Union, but which, at that point, might be recalibrated to protect against incursions from the US, as well. NATO has been mostly funded and perpetuated by the US until recently, however, so there’s a chance that something else would need to replace it, if the US is no longer providing nuclear deterrence as the ultimate whammy against a potential Russian invasion of its European neighbors. The UN has also indicated that they consider this operation to be a violation of international law, and have called it a dangerous precedent—because one nation capturing the leader of another nation, unilaterally, kind of negates the purpose of negotiations and the whole concept of international law. That kind of use of force is meant to be granted by the UN, not attempted secretively and outside the bounds of international processes for such things. All that said, the Trump administration seems to be leaning into the victory, gleefully talking about next-step potential targets, the most likely of which seem to be in Iran, a long-time US opponent, and a target of this administration last year, when the US attacked Iranian nuclear facilities alongside Israel. There are ongoing, very large and seemingly significant protests happening across Iran right now, so the US could see this as another opportunity to topple another unpopular authoritarian regime while also getting the chance to flex its military and intelligence capabilities at a

13 gen

•

14 min
6 GEN

Sports Betting

This week we talk about prediction markets, incentives, and gambling addiction. We also discuss insider trading, spot-fixing, and Gatorade. Recommended Book: The Kingdom, the Power, and the Glory by Tim Alberta Transcript Prediction markets are hundreds of years old, and have historically been used to determine the likelihood of something happening. In 1503, for instance, there was a market to determine who would become the next pope, and from the earliest days of commercial markets, there were associated prediction markets that were used to gauge how folks thought a given business would do during an upcoming economic quarter. The theory here is that while you can just ask people how well they think a political candidate will fare in an election or who they think will become the next pope, often their guesses, their assumptions, or their analysis will be swayed by things like political affiliation or maybe even what they think they’re meant to say—the popular papal candidate, for instance, or the non-obvious, asymmetric position on a big commercial enterprise that might help an analyst reinforce their brand as a contrarian. If you introduce money into the equation, though, forcing people to put down real currency on their suspicions and predictions, and give them the chance to earn money if they get things right, that will sometimes nudge these markets away from those other incentives, making the markets commercial enterprises of their own. It can shift the bias away from posturing and toward monetization, and that in turn, in theory at least, should make prediction markets more accurate because people will try to align themselves with the actual, real-deal outcome, rather than the popular—with their social tribe, at least—or compellingly unpopular view. This is the theory that underpins entities like Polymarket, Kalshi, Manifold Markets, and many other online prediction markets that have arisen over the past handful of years as regulations on these types of businesses have been eased, and as they’ve begun to establish themselves as credible players in the predicting-everything space. In politics in particular, these markets have semi-regularly shown themselves to be better gauges of who will actually win elections than conventional polls and surveys, and though their records are far from perfect and still heavily biased in some cases, such community-driven predictions from money-motivated markets are gaining credibility because of their capacity to incentivize people to put their money where their mouths are, and to try to profit from accurate preordination. The flip-side of these markets, and some might even say a built-in flaw with no obvious solution, is that they are rife with insider trading: people who are in the position to know things ahead of time making in some cases millions of dollars by placing big bets that, for them, aren’t bets at all, because they know what will or what is likely to happen. This seems to have occurred at least a few times with big political events in 2025, and it’s anticipated that it could become an even bigger issue in the future, especially for markets that use cryptocurrencies to manage payments, as those are even less likely than their fiat currency peers to keeps solid tabs on who’s actually behind these bets, and thus who might be trading on knowledge that they’re not supposed to be trading on. That said, it could be argued that such insider trading makes these markets even more accurate, eventually at least. And that points us toward another problem: the possibility that someone on the inside might look at a market and realize they can make a killing if they use their position, their power to sway these markets after placing a bet, giving them the ability to assure a payout by abusing their position—major events being influenced by the possibility of a community-funded payday for those in control. What I’d like to talk about today is the same general principle as it’s playing out in the sports world, and why the huge sums of money that are now sloshing around in the sports betting industry in the US are beginning to worry basically everyone, except the sports betting companies themselves. — In October of 2025, the head coach of the NBA basketball team, the Portland Trail Blazers, Chauncey Billups, Miami Heat player Terry Rozier, and former NBA player Damon Jones, and about 30 other people were arrested by the FBI due to their alleged illegal sports gambling activities. Rozier was already under investigation following unusual betting activity that was linked to his performance in a 2023 game—he was later cleared of wrongdoing, but the implication then and in this more recent instance is that he and those other folks who were rounded up by the FBI may have been involved in rigging things so they could get a big payoff on gambling markets. Similar things have been happening across the sports world, including a lifetime ban for Jontay Porter, a former Toronto Raptors player, who apparently gave confidential information to people who were placing bets on NBA games—he later pleaded guilty to conspiracy to commit wire fraud as a result of that investigation—and in November of 2025 two Major League Baseball players, both of them pitchers for the Cleveland Guardians, Emmanuel Clase and Luis Ortiz, were charged by federal prosecutors for allegedly rigging pitches to benefit people betting on those pitches; they’ve been charged with wire fraud and money laundering, and each could face up to 65 years in prison. And those are just a few of the many instances of game-rigging that have been alleged in recent years, the specifics of which vary, but the outcome is always to give someone an advantage in these markets, which are only recently broadly legal across the United States, and which thus allow folks with the right connections or some money to invest ahead of time to, for instance, pay a pitcher to throw an inning, or pay a coach to tell them who will be benched and when, so that they can make a big wager with less of a risk, or in some cases, no risk at all. One of the big issues here is that rather than simply being a which-team-will-win sort of thing, many of these bets are highly specific and granular, including what are called proposition or prop bets that allow folks to gamble on the number of strikeouts a pitcher will tally in a given inning and other very specific things. If a pitcher were to then place a bet, perhaps through an intermediary, on their own prop bet-related performance, they would stand a decent chance of tallying the right number of strikes and balls. They could also sell that information to someone else, taking a guaranteed payout in exchange for the foreknowledge they grant that gambler, who could then do what they want with the information, and then if they do well with it, they could pay that pitcher to do the same again in the future. This type of bet is called spot-fixing, and it’s seen across prediction markets, not just sports markets. Pitchers can fix an inning of a game, but poker players can also go all-in or fold a given number of times in a tournament, and the folks in charge of dumping Gatorade over the winning coach following a Super Bowl event can leak that color, based on their foreknowledge of the setup, to gamblers—these markets are sprawling and varied, and anyone in any position of power who can make decisions about such things, or who’s involved enough to leak information can do so at a profit, either themselves putting down money on spot-fixed prop bets, or selling that information to those who will themselves place a bet. The issue sports organizations in the US are now running into is that while they aligned themselves with sports gambling entities like DraftKings and FanDuel after these platforms were legalized in more states following the striking-down of a federal ban on such things in 2018—as I record this, they’re currently legal in 31 states, alongside Washington DC and Puerto Rico—and they’ve profited a fair bit from that, allowing these businesses to become sponsors, to slap their logos on everything, and to generally become interwoven with the leagues themselves; despite all that, they’ve also created a sports culture in which betting is ultra-common, and that means fans are no longer just fans, they’re putting down money on various possible sports-related outcomes. That means folks who were maybe previously die-hard fans of their local team may no longer just be disappointed when their team loses, they’ll be financially impacted, perhaps even devastated. And many athletes who play on these teams, in these leagues, are now suffering all kinds of abuse and threats from people who decided to put a lot of money on their performance, but who failed to win a game, or maybe even throw the exact right number of strikes and balls in a given inning. This points at two big issues with sports betting in the US right now. First is that there’s a lot of money splashing around in this space. An estimated $160-170 billion was wagered by US citizens in 2025 alone, generating about $16.4 billion in revenue for sportsbooks—the entities that take these sorts of bets. That’s likely a significant undercount, too, as more generalist prediction markets are also getting involved in the sports betting game, blending this type of gambling with other sorts of prediction markets, like those related to politics and international happenings, like war. And second, a lot of people are gambling a lot of money on sports stuff right now, and that’s becoming an issue. In October of 2025, a Pew Research poll found that 43% of US adults think legalized sports betting is bad for society, up from 34% in 2022, and 40% says it’s bad for sports, up from 33%. A whopping 22% of US adults say they personally bet money on sports in the past year, up from 19% in 2022, and 10%, one in ten American adult

6 gen

•

16 min
06/05/2025

Model Context Protocol

This week we talk about the Marshall Plan, standardization, and USB. We also discuss artificial intelligence, Anthropic, and protocols. Recommended Book: Fuzz by Mary Roach Transcript In the wake of WWII, the US government implemented the European Recovery Program, more commonly known as the Marshall Plan, to help Western Europe recover from a conflict that had devastated the afflicted countries’ populations, infrastructure, and economies. It kicked off in April of 1948, and though it was replaced by a successor program, the Mutual Security Act, just three years later in 1951—which was similar to the Marshall Plan, but which had a more militant, anti-communism bent, the idea being to keep the Soviets from expanding their influence across the continent and around the world—the general goal of both programs was similar: the US was in pretty good shape, post-war, and in fact by waiting to enter as long as it did, and by becoming the arsenal of the Allied side in the conflict, its economy was flourishing, its manufacturing base was all revved up and needed something to do with all the extra output capacity it had available, all the resources committed to producing hardware and food and so on, so by sharing these resources with allies, by basically just giving a bunch of money and assets and infrastructural necessities to these European governments, the US could get everybody on side, bulwarked against the Soviet Union’s counterinfluence, at a moment in which these governments were otherwise prone to that influence; because they were suffering and weaker than usual, and thus, if the Soviets came in with the right offer, or with enough guns, they could conceivably grab a lot of support and even territory. So it was considered to be in everyone’s best interest, those who wanted to keep the Soviet Union from expanding, at least, to get Europe back on its feet, posthaste. So this program, and its successor program, were highly influential during this period, and it’s generally considered to be one of the better things the US government has done for the world, as while there were clear anti-Soviet incentives at play, it was also a relatively hands-off, large-scale give-away that favorably compared with the Soviets’ more demanding and less generous version of the same. One interesting side effect of the Marshall Plan is that because US manufacturers were sending so much stuff to these foreign ports, their machines and screws and lumber used to rebuild entire cities across Europe, the types of machines and screws and lumber, which were the standard models of each in the US, but many of which were foreign to Europe at the time, became the de facto standard in some of these European cities, as well. Such standards aren’t always the best of all possible options, sometimes they stick around long past their period of ideal utility, and they don’t always stick, but the standards and protocols within an industry or technology do tend to shape that industry or technology’s trajectory for decades into the future, as has been the case with many Marshall Plan-era US standards that rapidly spread around the world as a result of these giveaways. And standards and protocols are what I’d like to talk about today. In particular a new protocol that seems primed to shape the path today’s AI tools are taking. — Today’s artificial intelligence, or AI, which is an ill-defined type of software that generally refers to applications capable of doing vaguely human-like things, like producing text and images, but also somewhat superhuman things, like working with large data-sets and bringing meaning to them, are developing rapidly, becoming more potent and capable seemingly every day. This period of AI development has been in the works for decades, and the technologies required to make the current batch of generative AI tools—the type that makes stuff based on libraries of training data, deriving patterns from that data and then coming up with new stuff based on the prompting of human users—were originally developed in the 1970s, but the transformer, which was a fresh approach to what’s called deep learning architectures, was first proposed in 2017 by a researcher at Google, and that led to the development of the generative pre-trained transformer, or GPT, in 2018. The average non-tech-world person probably started to hear about this generation of AI tools a few years later, maybe when the first transformer-based voice and image tools started popping up around the internet, mostly as novelties, or even more likely in late-2022 when OpenAI released the first version of ChatGPT, a generative AI system attached to a chatbot interface, which made these sorts of tools more accessible to the average person. Since then, there’s been a wave of investment and interest in AI tools, and we’ve reached a point where the seemingly obvious next-step is removing humans from the loop in more AI-related processes. What that means in practice is that while today these tools require human prompting for most of what they do—you have to ask an AI for a specific image, then ask it to refine that image in order to customize it for your intended use-case, for instance—it’s possible to have AI do more things on their own, working from broader instructions to refine their creations themselves over multiple steps and longer periods of time. So rather than chatting with an AI to come up with a marketing plan for your business, prompting it dozens or hundreds of times to refine the sales copy, the logo, the images for the website, the code for the website, and so on, you might tell an AI tool that you’re building a business that does X and ask it to spin up all the assets that you need. From there, the AI might research what a new business in that industry requires, make all the assets you need for it, go back and tweak all those assets based on feedback from other AI tools, and then deploy those assets for you on web hosting services, social media accounts, and the like. It’s possible that at some point these tools could become so capable in this regard that humans won’t need to be involved at all, even for the initial ideation. You could ask an AI what sorts of businesses make sense at the moment, and tell it to build you a dozen minimum viable products for those businesses, and then ask it to run those businesses for you—completely hands off, except for the expressing your wishes part, almost like you’re working with a digital genie. At the moment, components of that potential future are possible, but one of the main things standing in the way is that AI systems largely aren’t agentic enough, which in this context means they need a lot of hand-holding for things that a human being would be capable of doing, but which they largely, with rare exceptions, aren’t yet, and they often don’t have the permission or ability to interact with other tools required to do that kind of building—and that includes things like the ability to create a business account on Shopify, but also the ability to access and handle money, which would be required to set up business and bank accounts, to receive money from customers, and so on. This is changing at a rapid pace, and more companies are making their offerings accessible to specific AI tools; Shopify has deployed its own cluster of internal AI systems, for instance, meant to manage various aspects of a business its customers perch on its platform. What’s missing right now, though, is a unifying scaffolding that allows these services and assets and systems to all play nice with each other. And that’s the issue the Model Context Protocol is meant to address. The Model Context Protocol, or MCP, is a standard developed by AI company Anthropic, and it’s open and designed to be universal. The company intends for it to be the mycelium that connects large language model-based AI to all sorts of data and tools and other systems, a bit like the Hypertext Transfer Protocol, or HTTP, allows data on the web to be used and shared and processed, universally, in a standardized way, and to dip back into the world of physical objects, how standardized shipping containers make global trade a lot more efficient because everyone’s working with the same sized boxes, cargo vessels, and so on. The Universal Serial Bus standard, usually shorthanded as USB, is also a good comparison here, as the USB was introduced to replaced a bunch of other standards in the early days of personal computing, which varied by computer maker, and which made it difficult for those makers, plus those who developed accessories, to make their products accessible and inexpensive for end-users, as you might buy a mouse that doesn’t work with your specific computer hardware, or you might have a cable that fits in the hole on your computer, but doesn’t send the right amount of data, or provide the power you need. USB standards ensured that all devices had the same holes, and that a certain basic level of data and power transmission would be available. And while this standard has since fractured a bit, a period of many different types of USB leading to a lot of confusion, and the deployment of the USB C standard simplying things somewhat, but still being a bit confounding at times, as the same shaped plug may carry different amounts of data and power, despite all that, it has still made things a lot easier for both consumers and producers of electronic goods, as there are fewer plugs and charger types to purchase, and thus less waste, confusion, and so on. We’ve moved on from the wild west era of computer hardware connectivity into something less varied and thus, more predictable and interoperable. The MCP, if it’s successful, could go on to be something like the USB standard in that it would serve as a universal connector between various AI systems and all the things you might want those AI systems to access and u

06/05/2025

•

16 min
15/04/2025

Money Mules and Matchmakers

This week we talk about smishing, Huione, and scams. We also discuss money laundering, the Cambodian government, and Tether. Recommended Book: The Longevity Imperative by Andrew J. Scott Transcript The portmanteau ‘smishing’ combines SMS and phishing to refer to the practice of using text messages to trick the recipients of said messages into revealing information that allows scammers to access their victim’s accounts on various platforms. One common variation of smishing, which I’ve seen a lot recently, personally, are messages purportedly from toll road operators that tell the recipient they’ve got an unpaid toll, and they need to follow a link that’s provided in order to pay it. If the person receiving that message follows the instructions, they’ll tend to land on a webpage that’s convincing enough, which looks like the sort of site you might go to if you’re paying that kind of toll, online, and you enter your payment information and are then either immediately charged for this fake toll, or that information is used in some more cohesive manner—maybe the card is stolen, maybe it’s added to a larger collection of data they have on you which is then leveraged for a larger payout. This type of scam has become more common in recent years because of innovations deployed by what security researchers have called the Smishing Triad, which is a trio of mobile phishing groups operating out of China that seem to have refined their infrastructure and techniques so that messages they send via iMessage to iPhone users and RCS to Android users can bypass mobile phone networks and enjoy a nearly 100% delivery rate—which makes the name a little ironic, since these groups don’t use SMS to deliver these scam texts anymore, as those other methods of delivery are more reliable for such messages, these days. The big innovation introduced by these groups, though, beyond that deliverability, is the productization of mobile phishing, which basically means they’ve packaged up applications that allow their customers, which are usually smaller-time phishing groups and individuals, to share links to convincing-looking copies of Paypal, Mastercard, Stripe, and CitiGroup payment sites, among others, including individual banks, and that makes knee-jerk payments from the victims receiving these texts more likely, and less likely to set of alarm bells in the minds those receiving them, because they look like just normal payment sites. These pre-packaged scam assets also include regularly rotated web domains, which makes them less likely to trigger the recipient’s anti-scam software—their browser will be less likely to flag them as problematic, basically. And the Triad has hundreds of actual humans working desk jobs, worldwide, supporting their customer base, which again is a bunch of scammers that use this package of tools to try to steal money from their marks. All of this is enabled, in part, by clever emulation software that allows Triad customers to leverage legit and legit-seeming phone numbers from a computer or phone, those devices then sending out around 100 messages per second, per device, to phone numbers in the targeted region. They’re able to do this on a budget because of the efficiency of the software acquired from the Smishing Triad, and the Triad stays just ahead of regulators and law enforcement by rapidly iterating their offerings, which in turn does the same for all of their customers—which grants the benefits of a larger institution to all these individual and smaller scam groups. What I’d like to talk about today is another alleged backend for scammers, this one this more overt and public facing, and perhaps even more impactful because of its size and because of the nature of its offerings. — The Huione (hu-WAY-wahn) Group is a financial conglomerate primarily based in Cambodia, though it also has satellite offices in other countries, mostly in Southeast Asia. Folks use the entity’s QR codes to pay for stuff all around Cambodia, from restaurant tabs to hotel bills to supermarket tallies, and it offers normal banking stuff like checking and savings accounts, alongside things like escrow services and a cryptocurrency exchange. This is a company that buys billboards along major highways throughout the country and which has well-connected people in charge, including one of the Cambodian prime minister’s cousins, who is the director of a Huione company. In addition to its many legitimate offerings, though, Huione has also been accused to providing a range of gray and blackmarket products and services to folks who are doing skeevy but partially legal things, alongside wholly criminal enterprises, like a human trafficking outfit in Myanmar and folks running large smishing schemes in other parts of Southeast Asia. Huione’s primary offering for the criminal underworld though, is allegedly serving as a money laundering go-between. If you run a smishing scammer network, or a group that kidnaps people and sell them into various types of modern slavery in Myanmar, you may have trouble using the money you earn for these efforts because they’re off-book, blackmarket sorts of income. You need to clean, to launder that money to make it seem legitimate, so that you can put it in banks or otherwise use it to pay for things like you would with normal, non-illegally earned money. Money laundering matchmaker services maintain networks of what are called money mules, and these mules are sometimes individuals, and they’re sometimes shell companies with bank accounts or their own cryptocurrency wallets. If you’re scamming people out of their money, you might use this type of service to connect you with a money mule, and you provide that mule’s bank or crypto account information to your victim—so when you receive a scammy text message and follow it to completing, the bank your money is sent to will probably be that of a mule, not the person or group doing the scamming. So the victim transfers their money to that mule’s account, and the mule then moves said money from one account to another to another to another to another, eventually converting it into an asset like a cryptocurrency, once the path has been suitably muddled. They take their cut, which is often something like 15%, somewhere along the way, and you, their customer, the scammer, are handed neutralized, clean resources in the form of that cryptocurrency—which you can then convert into real money at some point—on the other end. An entity like Huione makes money by connecting scammers and other criminals with mules, but also by serving as a guarantor on these transactions. So this entity allegedly, via a network of Telegram channels it maintains, telegram being an anonymizing chat app similar to WhatsApp, it allows matchmakers to advertise on these channels, using thinly veiled language to promote their services, and Huione is able to make money selling ads to mules and other matchmakers who want to promote via these highly trafficked channels, one of which has more than 400,000 users—and they have many of these things, and that alone apparently brings in a fair bit of revenue, serving as a sort of hard-to-track Craigslist for this component of the scam economy. The guarantor component of this digital bazaar means that Huione holds the transactions between scammer and mules in escrow, just like any other escrow service: they take the money and hold it until the service has been completed, at which point they release it, taking a small cut for the service of ensuring that no one gets ripped off—except for the original victim of the scam, of course. The majority of these transactions are completed using Tether, which is a stablecoin that tries to peg its value to the US dollar, each token worth exactly one USD, rather than fluctuating like speculative crypto assets, like Bitcoin, and this allows everyone involved to maintain a veil of both feigned ignorance and anonymity, making it difficult to track who does what, how much money changes hands, and who gets paid and does the paying. This setup allows Huione to claim ignorance any time someone accuses them of doing illegal stuff: after all, they can’t possibly be responsible for what all the entities using their services are up to, right? All everything is just muddled and anonymized enough to grant seeming truthfulness to that claim of ignorance. Because of how all this is set up, most of what we know about this is the result of whistleblowing from insiders and leaked documents, alongside divulgences from security researchers who know how to get into these sorts of networks and who at times hack those involved in various ways. And it seems, based on those divulgences and other gleaned knowledge, that Huione’s money laundering services, alone, have been linked to nearly $27 billion in cryptocurrency transactions since 2021—though that could be a significant undercount because of the blurry nature of this industry and the entities involved with it. Thus far, Huione has never been targeted for sanctions by any government. Tether took action to freeze some of its accounts after law enforcement officials flagged them for criminal behavior, and Telegram has closed some of those illicit, matchmaking channels, but it’s easy enough to set up new versions of both, while the escrow subsidiary of Huione, previously called Huione Guarantee, denies any connection to these activities and even changed its name to Haowang Guarantee in October of 2024, though that denial seems to be public-facing only: the escrow-providing company continues to claim that the larger Huione Group is one of its strategic partners and shareholders. Huione also has its own matchmatching service, called Huione International Pay, which operates as a real-deal bank, but also does what all the other matchmakers do—it helps criminal enterprises shuffle their money around, taking a fee to provide them wit

15/04/2025

•

18 min
27 GEN

TikTok Deal

This week we talk about social networks, propaganda, and Oracle. We also discuss foreign adversaries, ByteDance, and X. Recommended Book: Rewiring Democracy by Bruce Schneier and Nathan E. Sanders Transcript In 2021, TikTok, a short-form video platform that’s ostensibly also a social network, though which leans heavily toward consuming content over socializing, was ranked the most popular website by internet services company Cloudflare, beating out all the other big tech players, including search engine juggernaut, Google. It was a neck and neck sort of thing, with Google taking the lead some days that year, but 2021 was definitely TikTok’s time to shine, as it was already popular with young people and was starting to become popular with the general public, of all ages and across a huge swathe of the planet. It even beat Facebook as the most popular social media website that year, despite, again, being mostly about consuming content rather than interacting—that was actually a prime motivator for Meta, which owns Facebook and Instagram, to redirect its own apps in a similar direction, shifting its focus from communication and interaction between users toward the creation of binge-able content, and feeding users more of that content in a feed optimized for time-losing levels of consumption. 2021 was also the first full year that TikTok was coming under scrutiny from the US government. In the preceding year, 2020, then first-term president Donald Trump said he was considering banning the app because it was becoming so popular, with young people in particular, and because it was owned by a Chinese company, ByteDance it represented a potential national security threat. So the idea was that because Chinese companies are forced, by their very nature, to do what the Chinese government tells them—that’s just how things work over there—and to do so on the down-low if that’s what the governments demands, and to lie about having to do what the government tells them to do, if the government tells them to thus lie, it doesn’t matter that ByteDance’s leadership swore up and down to the world that the company will never use its popularity, and the data it soaks up from all its users as a result of that popularity, to help the Chinese government, the Chinese military, or Chinese intelligence services. It of course will have to do that, and if it doesn’t, its leaders could be black-bagged and disappeared in the night—because again, that’s just how things work over there. So the Trump administration decided to make TikTok a sort of bogeyman, representing Chinese companies in general, and to some degree the presence of China in the US and throughout the Western world, and said, nope, we’re not gonna let this thing continue to operate over here. It’s worth remembering, too, that by 2021 the world was enmeshed in the COVID-19 pandemic, which originated in China, and which Trump and his administration were ardently attempting to tie to the Chinese government—calling Covid the Chinese Flu, and even worse things, as part of that effort. So this move against TikTok and its parent company, while based on genuine concerns about the ownership of the company and how and where the data being collected by said company is handled, it should also be seen as a political maneuver, allowing Trump, during the 2020 election run-up, to look like he was taking a big stand against a big foreign threat, China. What I’d like to talk about today is a deal that was proposed way back then by the Trump administration, as a potential way out for TikTok and ByteDance, allowing it to continue operating in the US despite threats to shut it down, now that said deal, or a version of it, seems to have finally come to fruition—and what we know about the shape of the resulting new, US-based version of TikTok. — On January 18, 2025, TikTok stopped worked in the US. It voluntarily suspended all services in the country in the lead-up to the implementation of the Protecting Americans from Foreign Adversary Controlled Applications Act, which was passed by the US congress and signed into law by then-president Joe Biden in April of 2024. This law gave social networking services controlled by ‘foreign adversaries’ 270 days, with the possibility of a 90-day extension, to divest themselves so that they’re no longer considered foreign adversary-owned. This law was almost exclusively aimed at TikTok, and the idea was that TikTok, in the US, would no longer be able to legally function following that deadline if it was still owned by China, which for the purposes of this law has been labeled a foreign adversary. ByteDance could keep TikTok in the US going if it sold a majority, controlling stake of its US-based assets to non-adversary owners, but otherwise it would have to shut down. Interestingly, though Trump was the original source of concerns about TikTok and its Chinese ownership during his first administration, when he stepped back into office in January 2025, he signed a new executive order that delayed the enforcement of this Biden-signed law, and then delayed it still-further, three more times after that, saying that he wanted to give American investors the time to negotiate controlling interest of US TikTok, rather than banning it. Those efforts eventually bore fruit in the shape of a new controlling entity called TikTok USDS Joint Venture LLC, which is made up of a bunch of non-Chinese investment entities, including US software behemoth Oracle, an Emirati investment firm called MGX, a US investment firm called Silver Lake, and a personal investment company owned by Michael Dell, the founder of Dell Technologies. There are other, smaller investors also involved, but the red thread that runs through almost all of them is that they’re big Trump supporters and funders, funneling a lot of money into Trump’s campaigns, and his family businesses. So six years after the initial legal salvo was fired at TikTok in the US, the local assets are now controlled by non-Chinese investors, though the original Chinese owner, ByteDance, still owns just under 20%, compared to about 15% apiece for Oracle, MGX, and Silver Lake. The new company’s board is majority-run by those investors, too, which means it’s majority-run by ardent Trump supporters. We don’t yet know what effect this will have on content within the app, but under full Chinese ownership, topics related to democracy, Tianamen Square, and the LGBTQ community, among others, were significantly downgraded in the algorithm, ensuring they were seldom shown to anyone, which in turn disincentivized content that those owners didn’t like while incentivizing content that was pro-China, and pro-Chinese government priorities. It’s considered to be likely, by analysts who watch these sorts of maneuverings, that the same will be true of this new entity, but for and against subject matter that the Trump administration is for and against. Which raises the possibility that the new US TikTok, while superficially the same as the previous US TikTok, will slowly go the way X, formerly Twitter, has gone under Elon Musk, which was dramatically pushed in a new direction under its own owner, focusing on his political and ideological priorities and punishing users who spoke against those priorities. TikTok could become more or less an extension of the Trump-verse, in other words, and could thus become something more akin to Trump’s own network, Truth Social, or other right-leaning and far-right social networks, like conservative YouTube-clone, Rumble, rather than something less ideological, or maybe I should say less overtly politically ideological, like Meta’s Facebook, Threads, and Instagram. Users have already noticed some changes to US TikTok after the change in ownership, though, including what sorts of data are collected. TikTok’s new privacy policy, which all users have to agree to before using the app, now that the platform has changed hands, says that TikTok will be using precise location tracking, keeping tabs on exactly where users are located via their device’s GPS. That’s compared to the app’s previous approximate location-tracking effort, which used SIM card and IP address data to understand general proximity—it still uses that data, too, but now, rather than knowing what neighborhood you’re probably in, it may also know what room in your house you’re scrolling from. The new US TikTok also tracks users’ interactions with AI tools, including their prompts, outputs, and metadata attached to said interactions, which includes details about where users are when they’re using such tools, and what time they used them. They also collect gobs of marketing data from outside sources, and based on the users’ activity within the app. So things you buy, websites and other apps you visit and use, and conversations you have will all be sucked up and agglomerated into a profile that’s then used to show you targeted advertising. This isn’t unique to US TikTok, but the company does seem to intend to make use of more such data, and to combine it with that other stuff it’s now collecting, to increase the price it can charge for ads, because they’ll be a lot more specifically targeted than before. Some users are beginning to comb through the new user agreement with a fine-toothed comb, noticing, in addition to those aforementioned major changes, that the company also reserves the right to collect information about your physical and mental health, to use identifying information in the videos and images you might share, and information gleaned from people and their identifying characteristics in images and videos, and to collect biometric data, which usually means eyes and faces and walking gate and things like that, to differentiate and track people across such content. They can keep tabs on your sex life, sexual orientation and gender, your drug usage, your ethnic and ra

27 gen

•

14 min
10 FEB

Grok's Scandals

This week we talk about OpenAI, nudify apps, and CSAM. We also discuss Elon Musk, SpaceX, and humanistic technology. Recommended Book: Who’s Afraid of Gender? by Judith Butler Transcript xAI is an American corporation that was founded in mid-2023 by Elon Musk, ostensibly in response to several things happening in the world and in the technology industry in particular. According to Musk, a “politically correct” artificial intelligence, especially a truly powerful, even generally intelligent one, which would be human or super-human-scale capable, would be dangerous, leading to systems like HAL 9000 from 2001: A Space Odyssey. He intended, in contrast, to create what he called a “maximally truth-seeking” AI that would be better at everything, including math and reasoning, than existing, competing models from the likes of OpenAI, Google, and Anthropic. The development of xAI was also seemingly a response to the direction of OpenAI in particular, as OpenAI was originally founded in 2015 as a non-profit by many of the people who now run OpenAI and competing models by competing companies, and current OpenAI CEO Sam Altman and Elon Musk were the co-chairs of the non-profit. Back then, Musk and Altman both said that their AI priorities revolved around the many safety issues associated with artificial general intelligence, including potentially existential ones. They wanted the development of AI to take a humanistic trajectory, and were keen to ensure that these systems aren’t hoarded by just a few elites and don’t make the continued development and existence of human civilization impossible. Many of those highfalutin ambitions seemed to either be backburnered or removed from OpenAI’s guiding tenets wholesale when the company experienced surprising success from its first publicly deployed ChatGPT model back in late-2022. That was the moment that most people first experienced large-language model-based AI tools, and it completely upended the tech industry in relatively short order. OpenAI had already started the process of shifting from a vanilla non-profit into a capped for-profit company in 2019, which limited profits to 100-times any investments it received, partly in order to attract more talent that would otherwise be unlikely to leave their comparably cushy jobs at the likes of Google and Facebook for the compensation a non-profit would be able to offer. OpenAI began partnering with Microsoft that same year, 2019, and that seemed to set them up for the staggering growth they experienced post-ChatGPT release. Part of Musk’s stated rationale for investing so heavily in xAI is that he provided tens of millions of dollars in seed funding to the still non-profit OpenAI between 2015 and 2018. He filed a lawsuits against the company after its transition, and when it started to become successful, post-ChatGPT, especially between 2024 and 2026, and has demanded more than $100 billion in compensation for that early investment. He also attempted to take over OpenAI in early 2025, launching a hostile bid with other investors to nab OpenAI for just under $100 billion. xAI, in other words, is meant to counter OpenAI and what it’s become. All of which could be seen as a genuine desire to keep OpenAI functioning as a non-profit arbiter of AGI development, serving as a lab and thinktank that would develop the guardrails necessary to keep these increasingly powerful and ubiquitous tools under control and working for the benefit of humanity, rather than against it. What’s happened since, within Musk’s own companies, would seem to call that assertion into question, though. And that’s what I’d like to talk about today: xAI, its chatbot Grok, and a tidal wave of abusive content it has created that’s led to lawsuits and bans from government entities around the world. — In November of 2023, an LLM-based chatbot called Grok, which is comparable in many ways to OpenAI’s LLM-based chabot, ChatGPT, was launched by Musk’s company xAI. Similar to ChatGPT, Grok is accessible by apps on Apple and Android devices, and can also be accessed on the web. Part of what makes its distinct, though, is that it’s also built into X, the social network formerly called Twitter which Musk purchased in late-2022. On X, Grok operates similar to a normal account, but one that other users can interact with, asking Grok about the legitimacy of things posted on the service, asking it normal chat-botty questions, and asking it to produce AI-generated media. Grok’s specific stances and biases have varied quite a lot since it was released, and in many cases it has defaulted to the data- and fact-based leanings of other chatbots: it will generally tell you what the Mayo clinic and other authorities say about vaccines and diseases, for instance, and will generally reference well-regarded news entities like the Associated Press when asked about international military conflicts. Musk’s increasingly strong political stances, which have trended more and more far right over the past decade, have come to influence many of Grok’s responses, however, at times causing it to go full Nazi, calling itself Mechahitler and saying all the horrible and offensive things you would expect a proud Nazi to say. At other times it has clearly been programmed to celebrate Elon Musk whenever possible, and in still others it has become immensely conspiratorial or anti-liberal or anti-other group of people. The conflicting personality types of this bot seems to be the result of Musk wanting to have a maximally truth-seeking AI, but then not liking the data- and fact-based truths that were provided, as they often conflicted with his own opinions and biases. He would then tell the programmers to force Grok to not care about antisemitism or skin color or whatever else, and it would overcorrect in the opposite direction, leading to several news cycles worth of scandal. This changes week by week and sometimes day by day, but Grok often calls out Musk as being authoritarian, a conspiracy theorist, and even a pedophile, and that has placed the Grok chatbot in an usual space amongst other, similar chatbots—sometimes serving as a useful check on misinformation and disinformation on the X social network, but sometimes becoming the most prominent producer of the same. Musk has also pushed for xAI to produce countervailing sources of truth from which Grok can find seeming data, the most prominent of which is Grokipedia, which Musk intended to be a less-woke version of Wikipedia, and which, perhaps expectedly, means that it’s a far-right rip off of Wikipedia that copies most articles verbatim, but then changes anything Musk doesn’t like, including anything that might support liberal political arguments, or anything that supports vaccines or trans people. In contrast, pseudoscience and scientific racism get a lot of positive coverage, as does the white genocide conspiracy theory, all of which are backed by either highly biased or completely made up sources—in both cases sources that Wikipedia editors would not accept. Given all that, what’s happened over the past few months maybe isn’t that surprising. In late 2025 and early 2026, it was announced that Grok had some new image-related features, including the ability for users to request that it modify images. Among other issues, this new tool allowed users to instruct Grok to place people, which for this audience especially meant women and children, in bikinis and in sexually explicit positions and scenarios. Grok isn’t the first LLM-based app to provide this sort of functionality: so called “nudify” apps have existed for ages, even before AI tools made that functionality simpler and cheaper to apply, and there have been a wave of new entrants in this field since the dawn of the ChatGPT era a few years ago. Grok is easily the biggest and most public example of this type of app, however, and despite the torrent of criticism and concern that rolled in following this feature’s deployment, Musk immediately came out in favor of said features, saying that his chatbot is edgier and better than others because it doesn’t have all the woke, pearl-clutching safeguards of other chatbots. After several governments weighed in on the matter, however, Grok started responding to requests to do these sorts of image edits with a message saying: “Image generation and editing are currently limited to paying subscribers. You can subscribe to unlock these features.” Which means users could still access these tools, but they would have to pay $8 per month and become a premium user in order to do so. That said, the AP was able to confirm that as of mid-January, free X users could still accomplish the same by using an Edit Image button that appears on all images posted to the site, instead of asking Grok directly. When asked about this issue by the press, xAI has auto-responded with the message “Legacy Media Lies.” The company has previously said it will remove illegal content and permanently suspend users who post and ask for such content, but these efforts have apparently not been fast or complete, and more governments have said they plan to take action on the matter, themselves, since this tool became widespread. Again, this sort of nonconsensual image manipulation has been a problem for a long, long time, made easier by the availability of digital tools like Photoshop, but not uncommon even before the personal computer and digital graphics revolution. These tools have made the production of such images a lot simpler and faster, though, and that’s put said tools in more hands, including those of teenagers, who have in worryingly large numbers taken to creating photorealistic naked and sexually explicit images of their mostly female classmates. Allowing all X users, or even just the subset that pays for the service to do the same at the click of a button or by asking a Chatbot to do it for them has increased

10 feb

•

16 min
3 FEB

Mother of All Deals

This week we talk about the European Union, India, and tariffs. We also discuss trade barriers, free trade, and dumping. Recommended Book: The Kill Chain by Christian Brose Transcript A free trade agreement, sometimes called a free trade treaty, is a law that reduces the cost and regulatory burden of trading between two or more states. There are many theories as to the ideal way to do international trade, with some economists and politicians positing that complete free and open trade is the way to go, because it allows goods and services to cross borders completely unencumbered, which in turn allows businesses in different countries to really lean into whatever they’re good at, selling their cars to countries that are less good at making cars, while that recipient country produces soy beans or computer chips or whatever they’re good at making, and sending those in the other direction, likewise unburdened by stiff tariffs or regulatory hurdles. Each country can thus produce the best product cheapest and sell it to the market where their products are in high-demand, while they, in turn, benefit from the same when it comes to other products and services. This theory leans on the idea that everyone is better off when everyone does what they’re best at, rather than trying to do everything—specialization. But those who oppose this conception of international trade argue that this creates and reinforces asymmetries between different nations and businesses: a country that’s really good at producing soybeans may be at a substantial disadvantage if the country that makes cars ever decides to go to war, because they won’t have the existing infrastructure to build tanks or drones or whatever else, while the country that specializes in computer chips might hold all the cards when it comes to generating economic pressure against its enemies or would-be enemies, because such chips are in everything these days, from military hardware to kitchen appliances. This also creates potential frailties for countries that specialize in, say, buggy whips, only to have a new technology like the automobile come around and put a significant chunk of their total economy out of business. This theory may also leave local businesses that don’t lean into a regional strength kind of in the lurch. If a country with a decent-sized automobile industry decides leaves their borders completely open to international competition, there’s a chance that could light a fire under those local producers, forcing them to become more competitive, but there’s also a chance it could collapse the market for local offerings—their cars might no longer be desirable, because the international stuff flooding across the borders from a nation that has heavily prioritized making cars are just so much better and cheaper, whether naturally or artificially, because of subsidies by that foreign government meant to help them take out international competition. This is why most nations have all sorts of tariffs, regulations, and other trade barriers erected between them and their trading partners, and why those trade barriers are ultra-specific, different for every single possible trade partner. The goal is to make international options less appealing by making them more expensive, or making it trickier for foreign competition to smoothly and quickly get their products on your shelves, while still making those things available in a volume that aligns with local consumer demands. And then ideally making it easier and cheaper for your stuff to get on their shelves. The negotiation of all this is massively complicated because Country A might want to favor their soybean farmers, who are an important voting bloc, and Country B might want to do the same for their car industry, because tax income from that industry is vital, and these two governments will thus do what they can to ensure their favored local industries and businesses have the biggest leg-up possible in as many foreign markets as possible, without giving away so much to their trade partners that they create worse situations for other industries and businesses (and the people who run them) on the home-front, as a consequence. What I’d like to talk about today is a recent, massive and potentially quite vital trade deal that was struck in early 2026, and what it might mean for global trade. — At the tail-end of January 2026, the European Commission announced that they had struck what they called “the mother of all deals” with India, this deal the culmination of two decades worth of negotiation, its tenets impacting about 2 billion people and around a quarter of the world’s total GDP. The agreement, as is the case with most such agreements, is fairly complex. But in essence it reduces or eliminates tariffs on 96.6% of all EU goods exported to India, which means about 4 billion euros of annual duties that would have otherwise been paid on European products in India will disappear—a savings for Indian consumers, and a boon for European producers whose products will now be cheaper in India. This is expected to be especially beneficial for European automakers like Volkswagen, Renault, and BMW, which have long been weighed down by a 110% tariff in India; that tariff will be reduced to as little as 10% on the first 250,000 vehicles sold, following this agreement. Lower priced vehicles will still face higher tariffs, to help protect India’s local carmakers, but electric vehicles will benefit from a five-year grace period, as India has been focusing on allowing as many cheap, renewable energy assets and infrastructure into the country as possible, regardless of where they come from. Tariffs on machinery, chemicals, and pharmaceuticals coming from the EU will be almost entirely eliminated, down from tariff rates of 44, 22, and 11%, respectively. Wine, which has long been tariffed at a rate of 150%, will be cut to between 20-30% for many varieties, and spirits from the EU coming into India will see 150% tariffs cut to 40%. On the other side of this deal, the EU will also open its market to Indian goods, reducing tariffs on about 99.5% of all such goods, including seafood, textiles, gems and jewelry, leather goods, plastic products, and toys. Several of these categories, like Indian seafood, textile-making, and other labor-intensive industries, have had a rough time of late, because of high US tariffs enforced by President Trump’s second administration, so this is being seen as a significant win for them in particular. Interestingly, while the reduction in trade barriers is substantial here, and the number of people and industries, and amount of money that’s involved is massive, this deal doesn’t include, and in some cases explicitly excludes, any agreements related to labor rights, climate commitment, or environmental standards. This means that while the European Union has thus far been pretty strict in terms of ensuring incoming products align with their policies and values regarding things like carbon emissions and ensuring goods aren’t produced by people laboring in slave-like conditions, this deal falls short of such enforcements, allowing India to operate with relative impunity, with regards to those issues, at least, and still sell with dramatically reduced barriers, on the European market. That’s a big deal, and is perhaps the biggest indicator of just how badly the EU wanted to make this deal work. The EU was also able to keep significant protections in place for important local sectors like beef and chicken, dairy, rice, and sugar—all industries in which India would have liked to compete in the EU, but which, because of those maintained barriers, they practically can’t. That would likely have been a feverishly negotiated topic, and it’s likely an indicator of how much India wanted this to work, too. On that note, both India and the EU were apparently especially interested in making this multi-decade deal work, now, because of increasing pressure from China on one side and the US on the other. China has been rerouting many of its cheap products that would have previously gone to the US market, elsewhere, engaging in what’s often called ‘dumping’ which slowly but surely puts businesses that produce comparable products at a profit in those local target markets out of business, at which point these Chinese companies can then ratchet up their prices and profits, operating without real competition. The EU and India have both been targeted by Chinese companies taking this approach, because they’re still producing at a feverish pace and because of US tariffs and the general unpredictability and irregularity of US policy overall under the second Trump administration, they’ve been firing that cheap product cannon more intensely at other large markets, instead—and India and the EU are the next two big markets in line right now, after the US and China. On the US side of things, those same tariffs have been hurting companies in both the EU and India that would otherwise been shipping their goods to the rich and spendy US market, and in many cases these tariffs have been fine-tuned to hurt important local industries as much as possible, because that’s one of Trump’s main negotiating tactics: lead with pain and then negotiate to take some of the pain away. This deal, then, serves multiple purposes in that it creates a valuable, newly polished trade relationship between a rich and powerful existing bloc and the newly most-populous country on the planet, which is also rapidly expanding economically and geopolitically. One last point to note, here, though, is that the European Union has been trying to create these sorts of mutually beneficial deals with non-US partners for a while, now, and the two most recent wins, trade deals with a South American trade bloc and with Indonesia, in early January 2026 and in September of 2025, respectively, have borne mixed results. The deal

3 feb

•

15 min
11/08/2020

Risk Perception

More information about Brain Lenses at brainlenses.com. BL supporters receive an additional episode of the show each week. Info about becoming a supporter at the above address. Read the written version of this episode: brainlenses.substack.com/p/risk-perception This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit brainlenses.substack.com/subscribe

11/08/2020

•

10 min
15/10/2024

Sleep and Memory

More information about Brain Lenses at brainlenses.com. Paid BL supporters receive an additional episode of the show each week. Read the written version of this episode: This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit brainlenses.substack.com/subscribe

15/10/2024

•

4 min

A portfolio of podcasts for curious, adventurous, passionate lifelong learners.