left

One fine day Richard Feynman left a restaurant. 50 years later we already know why better known bad than good unknown

by

In the late 1970s, the brilliant physicist Richard Feynman He went with his friend Ralph Leighton to eat at a Thai place named Indra in Glendale, California. Looking at the restaurant’s menu, Leighton couldn’t decide: should he order his usual favorite, ginger chicken, or try something new and perhaps better? Any other person would have responded in one way or another (“if you like it so much, you better insure” or something like “he who does not risk does not gain”). Richard Feynman, brilliant as he is, did something else: He started scribbling equations on a napkin. and he turned that into a mathematical problem that he not only detected, but solved. For some reason, the prodigious physicist never published that analysis, and his notes were left to Leighton. For years that story was forgotten, but 50 years later researchers from the universities of Oxford, New York and Princeton managed to rescue those notes and Feynman’s solution. And what that revealed was surprising. Rescuing Feynman’s restaurant problem The researchers explained in their study, published in PNAS (Proceedings of the National Academy of Sciences) that although Feynman had focused on what happened to the different dishes in the same restaurant, they They preferred to expand the problem: what happens when we are in city X, for example, on vacation, and we want to choose a restaurant. Richard Feynman’s handwritten notes on a restaurant napkin turned out to be a fascinating problem. Source: PNAS. Feynman’s restaurant problem is actually a variant of what is known as the optimal stopping problemto which also belongs the famous variant of secretary problemwhich gave rise to the 37% rule: When choosing from 100 options, one should try the first 37 to maximize the chances of choosing the best one. Then you can “settle” for that one, because it is difficult for there to be a better one among the rest. But we are digressing. Feynman’s original mathematical formula established an optimal policy based on a uniform distribution of quality. According to the physicist’s formulation, our quality bar is not static nor falls by chance, but decreases exponentially as the days available in our vacation calendar are exhausted. Thus, it usually happens that when we are at the beginning of our vacation, We usually demand absolute perfection in the chosen restaurant because the remaining time allows the risk to be amortized. In the end, however, that threshold of demand collapses and we settle for a decent restaurant. We move from the exploration phase – taking risks with new places (or dishes) – to exploitation – repeating places (or dishes) that we liked. The researchers wanted to test this mathematical model with a sample of 2,520 participants, and in doing so they detected a striking anomaly. During the first nights in a new city, participants explored massively, much more than mathematical logic itself advised. The researchers discovered that this phenomenon responded to the so-called “early exploration bonus” that fell rapidly as the days went by: if we have an opportunity to “get it right,” our brain shows tremendous psychological resistance to tying itself to a restaurant at the first opportunity. We prefer to continue trying other restaurants because we trust that we will find a better one. The four “gastronomic worlds” of the study: the behavior of the participants varied according to each distribution. Source: PNAS. But as the experiment went on, something else was discovered. Humans are not blind robots, but we calibrate the bar according to the city we visit. The experiment placed participants in four different “food worlds” in which the ratio of excellent restaurants to mediocre (or decent) ones varied. The data showed that the human brain is capable of diagnosing the type of “food world” it finds itself in just by trying three or four restaurants. From there, set the bar. Feynman mathematically intuited that the bar would lower exponentially as the return date approached, but the experiment revealed something different. Human beings reduce our level of demand linearly with respect to the proportion of days we have left on vacation. We are becoming less and less demanding and more “nostalgic”. This guarantees something important: that at least on the last nights we enjoy the “better bad known than good not known”, because that “bad known” will not be so bad after all: we have already experienced it. Fascinating. Image | SAP (edited with Magnific) In Xataka | Studying by heart seems like a good idea until you forget it. The Feynman method appeals to your understanding, not your memory.

A Renfe train was left without air conditioning at 40ºC. 1,056 days later, an affected person has just collected her money

by

How much would you fight for the refund of 31.90 euros? Macarena LE, a resident of a town in Teruel, is clear: three years, if necessary. And that is how long it has been behind Renfe for it to refund the price of a ticket between Zaragoza and Barcelona. The reason? The car’s air conditioning. Evident, due to the non-existent air conditioning of the car. What has happened? A Renfe user named Macarena LE, a member of Facua, has gotten the money back for a train ticket she took three years ago. The company had promised that it would do so with each and every one of the people who took that train, but the money never appeared in the user’s bank account. a little odyssey. Despite telling them that they would return the money to the passengers, the amount never arrived so after a few weeks, the affected person filed a claim, they explain in Facua. With no news about it after three months, he decided to put himself in the hands of the legal service of the consumer defense association. Once reaching this point, the association filed a new claim. To this complaint, Renfe responded that it gave the order to return the money but it still did not arrive. After another four months, Facua files a new complaint and Renfe assures him that it has tried to refund the money but that the procedure has been rejected on two occasions. At that moment, Renfe instructs Macarena LE to receive the money through a payment platform. This method also being impossible and the months accumulating, Facua sends a third claim to Renfe in which they simply ask it to deposit the money into the bank account of the affected person, attaching the document that proves its ownership. Three years after taking that train, the associate has ended up receiving her 31.90 euros. A trip… complicated. And, once seated, they assure in Facua that the conductor informed the passengers that the air conditioning was broken and that their travel money would be refunded in compensation for the inconvenience caused. But the journey between Zaragoza and Barcelona had an added problem: July 18 with departure at 3:52 p.m. and arrival at 5:20 p.m. That day maximum temperatures of 44 degrees were reported. Obviously, the heat must have been oppressive inside and the affected person claims that they had to drink water constantly to get through the trip in the best possible way. The expected quality. Due to the numerous claims filed by Facua, and beyond the fact that the money was not returned until three years later, Renfe never fought not to return the money. Instead they admitted that they would send the money back for “not having offered the service with the quality levels” expected. However, the fulfillment of the procedure itself does not seem to have had the expected quality either. And Facua pointed out to Renfe that they were failing to comply with article 21.3 of the Royal Legislative Decree 1/2007, of November 16according to which: “Employers must respond to the claims received in the shortest possible time and, in any case, within a maximum period of one month from the presentation of the claim.”. When Facua presented this second claim in which this regulatory text was recalled, Renfe had not given a response or paid the corresponding money for four months. What if it happens to me? That Renfe has returned the money to a passenger sets a precedent but does not automatically lead the company to return the money in the event of an air conditioning breakdown. In its rules, Renfe states that the traveler has the right to receive compensation for the cancellation of the trip, for its interruption, for a lack or deficiency in the service provided on board or for delay, but it makes it clear that such compensation will not be carried out if: The reason is an extraordinary circumstance unrelated to railway operation, such as extreme weather events (a storm that leaves a train without heating in winter, for example) or natural disasters. Nor if the breakdown or problem was caused by the behavior of the traveler. Nor if it is the damage caused by a third party outside the railway company that causes the incident (people on the track, copper theft, sabotage… etc.) That is to say, the mere absence of air conditioning does not automatically imply that the passenger’s money will be returned. What has happened will have to be taken into account to decide whether the company is obliged to act in this way. Photo | Nelson Silva In Xataka | The Madrid Cercanías have become a nest of problems and delays: their solution is new “megatrains”

Ukraine has been left without thousands of drones. An error classified them as electric cars, and the Treasury has fried them with taxes

by

During World War II, the United States Army created entire systems classification and emergency purchases because normal bureaucracy was too slow to keep up with the pace of war. Eight decades later, Ukraine has discovered the same problem from the opposite side. Drone warfare crashes into bureaucracy. Ukraine has been transforming the front into a war laboratory automated where ground drones have become essential to transport ammunition, evacuate wounded or attack Russian positions without exposing soldiers. The problem is that, while kyiv was trying to accelerate this military revolution, the bureaucracy has ended up mistakenly classifying these unmanned vehicles within the same tax category than electric cars. When an old exemption for EVs expired on January 1, drones began paying a 20% VAT. The result has been devastating: according to the industry, the army could have bought some 5,000 additional drones only in the first half of 2026 if that tax had not come into force. Thousands of drones lost at the worst moment. They counted on Insider that the impact has been especially serious because it has arrived at a critical phase of the war. Ukraine is increasingly relying on autonomous systems to compensate for human and material attrition against Russia, to the point that Zelensky claimed that his forces carried out more than 22,000 missions with ground drones in just three months. kyiv wanted to acquire 50,000 units this year, but the new VAT skyrocketed costs, froze public contracts and left manufacturers whole for months. no state orders. Some companies drastically reduced production to survive, while others tried to reclassify their robots as armored vehicles to avoid the tax burden. A trapped military industry. The chaos also reflects how the military technological revolution is advancing faster than the laws themselves. Ground drones were so new within European and Ukrainian commercial standards that they did not even there was a category clear to classify them. When a former tax exemption for electric vehicles expired, the system automatically absorbed these military robots into the same regulations. The Ministry of Defense suddenly found itself with insufficient budgets and paralyzed purchasing processes because, technically, essential weapons for the front had no longer been considered. exempt military equipment tax. Manufacturers like Tencorecreator of the popular TermIT dronethey spent up to five months without public contracts and had to survive thanks to volunteer organizations that directly supply military units. In a war economy where many companies literally live from order to order, three months without state purchases is equivalent to little less than a heart attack industrial. The big problem is not just making weapons. The episode reveals something deeper about the evolution of modern warfare. For years, drones, artificial intelligence and automation have been talked about as the future of combat, but Ukraine is discovering that the bottleneck is not always in the technology. Sometimes it is in the administrationin legislation or in bureaucratic systems designed for peacetime. Russia and Ukraine are immersed in a race of constant adaptation where every month counts and where losing half a year due to tax procedures can have direct effects on the front. The sector itself calculates that the tax exemption would save about 200 million dollarsa gigantic figure for an industry that still depends on precarious financing and accelerated production. The problem is that even if Parliament now corrects the law, the damage has already been done: delayed contracts, lost capacity and thousands of drones that never made it to the battlefield when they were needed most. The paradox of the war of the future. The story perfectly summarizes one of the great contradictions of this war. Ukraine has become the country that has integrated autonomous systems the fastest in real combat and has built an ecosystem with more than 280 companies and 550 models different from ground drones. However, that same ecosystem remains dependent on sluggish state structures, legacy regulations, and legal frameworks unable to keep pace with military innovation. While the front is filled with robots that transport ammunition, evacuate wounded or attack Russian trenches without a human driver, the State continued to administratively treat them as if they were simple electric cars. The irony could not be more brutal: one of the most technologically advanced wars of the century lost thousands of combat machines not due to lack of industrial capacity or due to Russian attacks, but because the Treasury decided to apply the same tax treatment than to a civil electric vehicle. Image | x In Xataka | A Ukrainian stork has managed to outwit a Russian drone in flight. The video is the best clue about who will win the war In Xataka | Ukraine has been terrorizing Russian soldiers with its heavy drones for years. Now they are literally giving it back.

China has more solid-state battery patents than anyone else and still fears being left behind for one reason: Japan

by

What China is leading the energy and mobility transition What we are witnessing does not take anyone by surprise at this point. However, not all fish are sold, and in energy storage we are going to witness a significant evolution with the arrival of solid state batteriesa type of battery that we have been talking about for years. Just like they count From CarNewsChina, the country dominates the volume of research and records on solid-state batteries, but be careful because that leadership on paper does not guarantee winning the commercial race. And it is that a new analysis of the Xinhua agency recognizes that the United States, Europe, Japan and South Korea are moving forward with more industrial coordination and better international deployment of patents, just when the technology enters a decisive phase for its commercialization. Why it matters. Solid-state batteries are considered the next big leap from current lithium-ion batteries. These promise more energy density, faster charges, greater security and longer lifespan. They not only affect the electric car, but also humanoid robotseVTOL (vertical take-off aircraft), consumer electronics and stationary storage. Basically, whoever controls the technology and, above all, its manufacturing at scale, will set the pace of mobility and energy in the next decade. Patent war. China accounts for around 35% of the world market for solid-state patents and 39% of those related to electrolytes, the largest global share, according to share from CarNewsChina. Scientific publications have gone from 21 articles in 2015 to 562 in 2023, with institutions such as the Chinese Academy of Sciences or Tsinghua University leading advances in the engineering of the solid-solid interface, which for years has been one of the great bottlenecks. On the other hand, Japan continues to be the leading technological source with about 37% of global requests, compared to 30% from China. Japan is ahead. The problem in China is not quantity, but the concentration and quality of your strategy. Among the 30 most relevant institutions in the world in solid state and electrolyte patents, 17 are Japanese7 Chinese, 5 South Korean and only 1 European. The top ten positions are entirely Japanese or Korean. Toyota, alone, accumulates around 40% of all intellectual property in the sector. Added to this is a structural weakness, since Chinese companies register many fewer international patents than their Japanese and South Korean rivals, who shield their technology in the United States, Europe, India and Southeast Asia. The companies that move the board. CATL, BYD and SVOLT are leading the latest phase of this technology. And only in 2023 will Chinese companies filed more than 500 patent applications. Gotion High-tech The design of a 2 GWh line for totally solid batteries has already been finalized and another 0.2 GWh pilot line is operating with tests on vehicles. Ganfeng Lithium, backed by Changan, claims to have reached 1,100 cycles in a 400 Wh/kg cell and aims for 500 Wh/kg in production. On the other hand, Chinese researchers have also shown a prototype of 451.5 Wh/kg capable of charging in three minutes. CATL, for its part, is patenting lithium compounds with fluorine and sulfur electrolytes to improve fast charging and thermal stability. Deadlines. own report Xinhua places the start of production in small series around 2027 and broader commercialization around 2030. The industry continues to work in parallel on three electrolyte routes (sulfide, oxide and polymer) without any having won yet. Furthermore, according to the media, there are still challenges to overcome, including the formation of lithium dendrites, ionic transport mechanisms, solid-solid interface engineering or cell failure modes. And now what. China is preparing to industrialize what it currently masters in the laboratory. And its first national standard on solid state batteries (“Terms and Classification”) is under public consultation and proposes to differentiate between liquid, solid-liquid hybrid and totally solid cells. For now, the country dominates in terms of volume of papers and research, but it is clear that real dominance will come from manufacturers who first resolve large-scale production, cost, durability and safety. And let’s be honest, China has an advantage, especially with CATL and BYD controlling much of the world’s battery sharebut in the field of solid-state batteries there is still play. Cover image | Michael Fousert In Xataka | The EU no longer knows what to do to stop its car manufacturers from buying parts from China. So he’s going to force them

If you have half a million euros left over, you can buy it

by

The Chinese company Unitree Robotics just presented the GD01, a manned robot that combines bipedal locomotion with movement on four limbs. Wow, a mecha that seems straight out of the movies but already has a price and production date. It already exists and can be purchased. According to Unitree, the GD01 is a high-strength alloy machine that weighs about 500 kilos with a pilot on board. To control it, simply place yourself in the cabin that incorporates the torso. Its starting price is 3.9 million yuan (about 538,000 euros at the current exchange rate). The company defines it as the world’s first mass-produced transformable mecha. What it can do. In the video published The company shows the GD01 walking upright on two legs and knocking down a brick wall with one hand. Next, the robot reconfigures its chassis and begins to move supported by four limbs, literally like a Transformers. China and robotics. The GD01 comes at a time when Chinese robotics companies are gaining ground notably compared to its American competitors, driven by lower production costs and greater manufacturing speed. According to consulting firm Omdia, Chinese companies accounted for almost 90% of global sales of humanoid robots in 2025. Unitree alone shipped more than 5,500 units last year, according to share SCMP, compared to the approximately 150 units shipped by each of the large American firms such as Tesla, Figure AI or Agility Robotics. The price gap. Unitree’s humanoid entry robot, the R1it costs around 5,500 euros to change. Its Chinese rival AgiBot has a simplified model for about 12,800 euros. And on the other hand, Elon Musk has estimated that the Tesla Optimus It could cost between $20,000 and $30,000 in the future. The GD01 is a different bet from the rest, especially to provide maneuverability in industrial environments. Unitree is in full expansion. The company already sells its R1 and G1 humanoid robots, as well as the Go2 robot dog, in international markets such as North America, Europe and Japan through AliExpress. Their robots have begun to appear in all kinds of environments and events (in fact we brought one in the last Xataka Awards). In March, Unitree also requested to go public in the Chinese market, with a financing plan of about 4.2 billion yuan, of which 85% would go to research and development. The question that remains in the air. The GD01 is, for now, a demonstration of technological capacity and a declaration of intent. It is also a really eye-catching product, which is precisely what the company is looking for: notoriety. It is certainly achieving it, although it is still up in the air whether its technological capabilities exceed those currently found in the industrial environments for which it is intended. Now, what’s cool is cool. In Xataka | We had already assumed that AI and robots were superior to humans at chess. Now they are also good at ping-pong

Neighbors in Chile tried to stop an Amazon data center. Justice has left a clear message with its decision

by

Artificial intelligence has been part of our lives for a long time, often almost without us stopping to think about what is behind it. We use it as if everything were happening in an invisible layer: models, algorithms and, perhaps, servers in some remote location. But we can also look at it from another perspective. The infrastructure that supports that world is very real: it has a location, consumes resources, requires permits, involves enormous investments, and can also alter the environment of those who live nearby. That is one of the great debates that is beginning to accompany the rise of AI: the cloud also has neighbors. They lost the case. A specific case leads us to Huechurabanorth of Santiago de Chile, where Amazon plans to build a data center. The initiative had received a favorable Environmental Qualification Resolution in July 2024, but not everyone was convinced that the project had been evaluated accordingly. That concern reached the judicial route through a claim presented by Patricio Hernández Valenzuelaa resident of the area, and the Second Environmental Court resolved on April 9, 2026 to reject ita decision that leaves the data center in a position to move forward. A very specific concern. Hernández questioned whether the environmental evaluation of the project had not adequately taken into account a possible high voltage line that, according to his approach, would be necessary to power the data center. The criticism was not minor: if both infrastructures were linked, they had to be analyzed together. For residents, not doing so meant leaving relevant impacts on the environment out of the analysis. The key to the failure. The court’s reasoning involves clearly separating both pieces. The ruling concludes that the data center and the eventual high-voltage line cannot be considered to form a single initiative, among other things because the Amazon project does not include that infrastructure as part of its design. Furthermore, the planned electricity supply does not depend on its own installation, but on the network managed by third parties, which reinforces the idea that these are different projects. Without joint evaluation. Once the existence of a project unit has been ruled out, the court concludes that an integrated environmental assessment is not appropriate. The sentence explicitly states it: “it has been proven that between both initiatives there is no relationship of functional interdependence that conditions their execution.” This nuance is key, because it implies that the data center can operate using the available electrical infrastructure, without the need to subject its viability to a future high voltage line which, in any case, would have to be evaluated separately if it were to be considered. Beyond the legal debate. The Amazon project has very specific dimensions on paper. The data storage center in Huechuraba is designed to operate for 30 years, with an estimated investment of 205 million dollars. It would be built on an area of ​​10.9 hectares, with a construction of 21,350.07 square meters, in the street of Américo Vespucio 1055. From the company, collects Reutershave pointed out that the design of the infrastructure focuses on minimizing energy and water consumption, and maintains that the plan met environmental requirements. Chile as a hub. The Huechuraba project is not an isolated initiative within Amazon’s strategy. Amazon Web Services has proposed an investment of more than 4,000 million dollars in Chile over 15 years to build, operate and maintain its infrastructure in the country. The idea is to turn Santiago into its third major center in Latin America, after São Paulo and the central region of Mexico. Factors such as connectivity through fiber optic cables are added to this context. The concern of those who live nearby. Beyond the investment and digital infrastructure they promise, data centers are often accompanied by very specific concerns: high electricity consumption, use of water for cooling, heat or noise generation, and their fit into environments that, in many cases, have environmental or community value. Google did not have the same path. The case of Amazon is not the only one that has gone through this type of debate in Chile. Google had obtained initial approval in 2020 to build a $200 million data center in Cerrillos, southwest of Santiago. However, the project’s journey was different. In February 2024, the Second Environmental Court decided to partially reverse that permissionand months later the company announced that it would not continue with the initiative as it had originally been proposed, opting to start a new process from scratch for a project in the same location, but with a redesign based on air cooling. Electricity enters the scene. If we broaden the focus, the debate is not limited to a specific project, but to the system’s capacity to absorb this type of infrastructure. A Systep reportpublished on September 23, 2025 with data from the National Electrical Coordinator, indicated that, taking 2025 as a starting point, the electrical demand of data centers in Chile could increase by 270% in five years. The same projection places this consumption at around 1,207 MW in 2030. These figures help to understand why the energy issue has become one of the central axes when talking about the expansion of the cloud and AI. Images | Xataka with Nano Banana In Xataka | In 2024, Big Tech spent absurd amounts of money on AI. In 2025, they managed to spend 77% more

China generated half of the digital viewing of the last World Cup. There is one month left until 2026 and it is still not clear if they will issue it

by

Less than five weeks before the whistle that will kick off the opening match of this year’s World Cup, FIFA has signed broadcast contracts with more than 175 countries. China and India, with almost three billion inhabitants, are not among them. It is the unpleasant fruit of a price war over broadcast rights that pits the largest football organization in the world against the two most populated markets on the planet. What is at stake. The mbiggest World Cup in historywhich is said soon: 48 teams, 104 matches to be played in USACanada and Mexico between June 11 and July 19. FIFA is selling it as the most watched and broadcast event of all time. If they manage to resolve the conflicts with the two countries with the largest number of inhabitants on the planet, of course. According to data from FIFA itselfChina generated 49.8% of all viewing hours on digital and social platforms during the Qatar 2022 World Cup. Half of global digital consumption. More: India added 32 million digital viewers in the final alone. They are two very important markets that should not be ignored. Why is this happening? Part of the explanation is in the schedules. The tournament is held in North America, which means that the highest-rated matches will start at 3:00 a.m. in Beijing and Shanghai, and at 12:30 a.m. in New Delhi. These are schedules that destroy the advertising market: there is not enough audience beyond the fans, and advertisers are reluctant to pay the very high rates for the events. And without substantial advertising revenue, networks cannot support the tens of millions of dollars that broadcasts cost. India: bidding war. JioStar, India’s largest media conglomerate (the result of the merger between Viacom18 and Disney Star), even offered $20 million for the rights. And FIFA rejected the offer: it wanted 100 million dollars for a package that would also include the rights to the 2030 World Cup. According to local mediaFIFA would have lowered its price to around 35 million, although the negotiation is still not closed. China: crazy prices. ApparentlyFIFA would have demanded between 250 and 300 million dollars for the rights in the Chinese market, a figure that CCTV (the only broadcaster authorized by law to negotiate these rights) would not be willing to even remotely match. Its budget is around 60-80 million dollars, according to the same sources. FIFA may be willing to go down to between 120 and 150 million, but it is still double what CCTV wants to pay. On social networks, fans protest the difference in numbers between China and India. They are their traditions and they must be respected. CCTV has broadcast the World Cup without missing a single edition since Argentina 1978. Previously, agreements were closed with enough notice to launch promotional campaigns and attract sponsors, but this time there is no agreement, and the tournament starts in five weeks. For example, In the 2018 and 2022 World Cups, CCTV had the rights closed months in advance. And to this is added an extra problem: journalists from the country have had difficulties obtaining visas to cover the World Cup, which would reduce the quality of the broadcasts and, consequently, weaken the attractiveness for Chinese sponsors (which, as is easy to imagine, are among the main sponsors of the tournament). High tension. What we have right now are two millionaire forces pulling the rope in different directions: both want the highest profitability, knowing that time is an absolutely essential variable, because each week without a signed deal is equivalent to advertising and sponsorships that disappear. Not to mention the exasperation of millions of fans, who are now turning Asia into a sea of ​​nail-biting fans. And not in the penalty shootout, precisely. In Xataka | You will only be able to get to the World Cup stadiums in the USA and Mexico by car. And they are going to charge you 300 dollars to park it

Anthropic has just left behind Claude’s biggest burden. He has achieved this after sealing an alliance with Elon Musk’s SpaceX

by

There are few things more frustrating than finding a tool that fits almost exactly what we need and discovering, just as we’re starting to get the most out of it, that we can’t keep using it at the same rate. Claude It has earned a prominent place among those who use artificial intelligence to program, analyze documents or work with demanding tasks, but it has also drawn a very specific complaint: its limits of use. We are not talking about a minor annoyance, but rather a friction capable of breaking the workflow. Anthropic has decided to attack the problem. The company led by Dario Amodei announced a rise of the limits of Claude Code and the Claude API, relying on a new alliance with SpaceXAI. The pact will give it access to Colossus 1, an infrastructure that Anthropic presents as a way to directly improve the experience of its most intensive users. The promise, for now, is clear: more room to use Claude without demand taking its toll so quickly. The tension with limits. The adjustment that helps understand this news came a few weeks earlier. Anthropic recently modified their time limits to better manage demand during peak hours. In practice, this meant that five-hour sessions could be consumed before those actual five hours had passed if the use occurred during peak periods. The change especially affected those who made more intense use of Claude. More room to use Claude. Anthropic specifies the improvement in three changes that, according to the company, take effect immediately. The first is the doubling of Claude Code’s five-hour limits for Pro, Max, Team, and Enterprise plans per seat. The second is the removal of the peak limit reduction for Claude Code on Pro and Max accounts. The third affects the API: Anthropic says it has considerably raised the usage limits for Claude Opus models, although the exact scope depends on the limits table published by the company itself. Colossus muscle 1. The agreement with SpaceXAI is the most striking piece of the announcement because Anthropic ensures that it will be able to use all the computing capacity of the Colossus 1 data center. According to the company, that means more than 300 megawatts of new capacity and more than 220,000 NVIDIA GPUs that will be available within a month. SpaceXAI also details that the cluster includes deployments of H100, H200 and GB200 accelerators. The transformation continues. SpaceXAI does not appear in this agreement as simply a new label within the SpaceX ecosystem. The context, Elon Musk noted that “xAI will be dissolved as an independent company” and that its artificial intelligence products will be integrated under SpaceXAI. The phrase helps understand why Anthropic is talking about this brand when explaining its new access to computing power. Of course, to avoid confusion, what Anthropic announced is not a purchase or a merger, but rather an agreement to use AI infrastructure. It is not an isolated agreement. Anthropic also wanted to frame the alliance with SpaceXAI within a much broader capability strategy. The company recalls an agreement of up to 5 GW with Amazon, which includes almost 1 GW of new capacity by the end of 2026, and another 5 GW pact with Google and Broadcom that will begin to come into operation in 2027. To this it adds a strategic alliance with Microsoft and NVIDIA, with $30 billion of capacity in Azure, and an investment of $50 billion in AI infrastructure in the United States with Fluidstack. The most futuristic part. The agreement also includes a much more speculative derivative. Anthropic says that as part of the pact, it has expressed interest in collaborating with SpaceXAI to develop several gigawatts of orbital computing capacity. SpaceXAI presents it as a possible answer to the pressure that AI is putting on energy, land and cooling on the ground, but for now we are far from something tangible. Of course, this route would only make sense if important engineering challenges are overcome first. The real challenge. Anthropic has put on the table a direct answer to one of the big complaints surrounding Claude, although the most important part is still missing: checking how it feels in real use. SpaceXAI’s new limits and additional capacity seem to point in the right direction for those who work intensively with these services. The improvement, therefore, opens a new phase: that of checking if Claude can offer more margin without its users encountering the same wall again too soon. Images | Xataka with Nano Banana In Xataka | The “token economy” is broken: flat AI programming fees are mathematically unsustainable

US sanctions have left its current market share at 0%

by

In just two years Nvidia has gone from leading the GPU market to artificial intelligence (IA) in China to not sell practically anything in this country. In fact, Jensen Huang, the CEO of this company, has confirmed just a few hours ago now Its market share in this Asian nation is 0%. This dramatic scenario for Nvidia is the result of two decisive strategies of the governments of China and the United States. Early October 2024 the Chinese Administration sent to its AI companies a recommendation asking them to use chips produced in China as much as possible. Ten months later this recommendation became a requirement. And the Chinese Government is already forcing state-owned data centers throughout the country to use at least 50% Chinese integrated circuits in their servers. The Administration led by Xi Jinping has made this decision because it can afford it. And it is that It already has three very clear alternatives to Nvidia: Cambricon Technologies, Huawei and Moore Threads. This panorama has led Jensen Huang to decide to openly criticize the US export policy. AND has done it on several occasions during the last few years. The head of Nvidia is not enough to be the most attractive option for his Chinese customers; It also has to deal with the decisions of the US Government. US sanctions on China are destroying Nvidia Jensen Huang holds that the US will not protect its technological hegemony by blocking AI chip exports; According to this executive, what Donald Trump’s Government must do is ensure that the American AI ecosystem is dominant worldwide. The current scenario proves him right, but at the moment nothing indicates that the US Administration is going to bet on its strategy. At least not in a consistent way from a practical point of view. And the US Department of Commerce does not give the slightest respite to American AI chip designers. When these companies receive an order from one of their Chinese clients must apply for an export license to this government entity and indicate which GPU they intend to send to China, their specifications and which client is going to use them, among other relevant information. The Office of Industry and Security is also responsible for carrying out investigations into the tariffs deployed by the Administration led by Donald Trump. Once the bureaucracy has been put in place, Department of Commerce technicians analyze export applications in the framework established by current regulation and approve or deny the sale of integrated circuits to China. This is the usual procedure, so there is nothing new up to this point. However, Nvidia, AMD and other American AI chip designers face a very serious problem: the Commerce Department takes several months to process their export licenses. The staffing of the Department of Commerce has been drastically reduced in recent months, and in the current context this scenario represents a very serious problem. The Industry and Security Office of this entity is not only responsible for processing export licenses linked to AI chips; is also responsible for carrying out tariff investigations deployed by the Administration led by Donald Trump. And with fewer personnel than in 2024 and 2025 it cannot cope. According to Bloombergthe Office of Industry and Security has lost 101 employees in recent months, which represents a 19% reduction in staff compared to what it had in 2024. Curiously, the staff specifically dedicated to developing standards linked to the semiconductor industry and reviewing applications for export licenses has decreased by 20%, although at the moment it has not emerged what is the reason for this staff drain. Be that as it may, during 2025 the Office of Industry and Security took an average of 76 days to resolve export requests, but this period is increasing in 2026. Very bad news for Nvidia and AMD. Image | NVIDIA More information | Tom’s Hardware In Xataka | The US remains committed to stopping China. Now it has targeted the second largest Chinese chip manufacturer

“Left click, right click”, this is how the AI ​​decides an attack in war. China, Russia and the US need fewer and fewer humans

by

A group of Google engineers signed an internal letter to protest against a project in which your own software was being used by the Pentagon, sparking an unprecedented debate within the company about how far the technology they had created should go. Since then, almost 10 years have passed, an “eternity” with the implementation of AI. The war accelerates… without humans. They counted last week in The New York Times that modern warfare is entering a phase in which human intervention is no longer the center of decision-making, but rather an almost symbolic step within processes dominated by algorithmswhere artificial intelligence systems identify targets, recommend attacks and generate complete plans in a matter of seconds. Programs like Project Maventoday developed by Palantir and integrated with models like Anthropic’sshow the extent to which the decision chain has been compressed: satellite images, drone data and intercepted signals are automatically processed to generate target lists and attack solutions, reducing human intervention to something as simple as selecting options on the screen, in the words of Pentagon officialsit is as simple as clicking “Left click, right click”. Powers in the same race. Because at the center of this transformation are the United States, China and Russia, competing to lead a new arms race based on autonomous systems capable of operating without direct intervention. In China, for example, the development of coordinated drone swarms by artificial intelligence and capable platforms to operate alongside fighters manned reflects a commitment to scale and automation. Meanwhile, in Russia they are betting on systems like the Lancet droneswhich evolve towards capabilitiesand autonomous selection of objectives. For its part, the United States is trying to close the gap by encouraging companies like Anduril to speed up production of autonomous drones, in a race where the speed of development is almost as important as the technology itself. The Chinese WZ-8 drone Ukraine as a turning point. How have we been countingthe war in Ukraine has been the turning point that has turned these technologies on real tools combat, demonstrating that relatively simple systems can evolve rapidly towards semi-autonomous capabilities and changing the balance on the battlefield. Adapted commercial drones, unmanned vessels and data analysis systems have allowed resist a superior adversary, while Russia has responded incorporating automation progressive in their own systems. As pointed out analyst Michael Horowitz, “the battlefield in Ukraine has served as a laboratory for the world,” accelerating a transition that is no longer experimental, but operational. Silicon Valley at war. Unlike previous arms races, the Times I remembered that the role does not fall solely on the States, but also in technology companies and start-ups that are redefining military development. Here are companies like Google that initially participated in projects like Maven before withdrawing due to internal pressures, while others like Palantir or Anduril have occupied that space with a more vision aligned with the defense. In China, the “civil-military fusion” model directly integrates to private companies in the development of military systems, while in the West attempts are made to replicate that dynamism with million-dollar investments and growing collaboration between Silicon Valley and the Pentagon. Algorithms against algorithms. The result is a form of war in which the confrontation is no longer only between armies, but between automated systems that operate at speeds impossible for humans, a scenario that we have counted where drones launch drones to take on other drones and sensor networks connect globally to execute real time attacks. Projects like the Chinese attempt to replicate networks similar to the Joint Fires Network American forces reflect this trend toward an interconnected war, one where a sensor at one point on the planet can trigger an attack on another without direct intervention. At this point, superiority no longer depends solely on the quality of weapons, but on the ability to integrate data, process it and act faster than the adversary. Uncontrolled speed. There is no doubt, this acceleration carries risks that worry even those who pushed these systems, as automation can trigger military responses before humans can intervene or fully understand the situation. Studies such as that of RAND Corporationworks that have shown scenarios in which autonomous systems inadvertently escalate conflicts, while experts warn of a possible “escalation spiral” driven by the decision speed of machines. As recognized General Jack Shanahan, promoter of Maven, the reality is that there is a danger of deploying “untested, insecure and poorly understood” systems in a context of competition where each actor fears being left behind. Less humans, more automation. Thus, the panorama that is drawn is that of a war every time more automatedwhere human intervention is progressively reduced and critical decisions are delegated in artificial intelligence systems capable of analyzing, deciding and acting in seconds, something very different which is do it “well”. From autonomous drones to target analysis platforms, through global combat networks, the trend seems clear, that of a war of the immediate future that will be decided less in offices and more in algorithmsin an unstable and certainly chilling balance, because we are talking about technological speed being on track to surpass the human capacity to control it in the middle of a war. Image | StockVault, Infinity 0 In Xataka | Russia is no longer surrendering to Ukrainian soldiers, but to machines: the rules of war are being redefined In Xataka | China was the power that launched drones. Now he has realized his danger with a decision: close the sky to them

Log In

Forgot password?

Forgot password?

Enter your account data and we will send you a link to reset your password.

Your password reset link appears to be invalid or expired.

Log in

Privacy Policy

Add to Collection

No Collections

Here you'll find all collections you've created before.