training

Science confirms how many minutes of weight training per week reduce the risk of mortality

by

For decades, when we thought about doing physical exercise, our minds almost automatically went to get older. cardiovascular activity. Running, swimming or cycling have been star recommendations to keep the heart healthy and extend life expectancy. Or at least live with a better quality of life. However, little by little we are normalizing the need to prioritize strength exercises at any age. How long. This is one of the big questions that anyone who needs to quantify the amount of exercise they do per day can ask themselves. There are clear recommendations, such as walk one hour a day at a brisk pacebut in strength we were quite orphaned. Now a new and monumental analysis has come to put exact figures on what until now were general recommendations, establishing a precise time window to maximize our years of life. What has been seen. The finding comes from a large observational study which has had 147,374 participants and exhaustive follow-up that has extended up to 30 years. Its good results have been published in the magazine British Journal of Sports Medicine. And when it comes to lifting weights or doing resistance exercises, intuition could dictate that “the more, the better”, but human physiology provides more limited metrics. The study data found that spending between 90 and 119 minutes weekly in resistance training routines was directly associated with lower overall mortality. In other words, spending between an hour and a half and two hours a week working our muscles is linked to a lower risk of dying from any cause. We have to be adjusted. What is truly revealing about this study lies in what happens when those 120 minutes are exceeded of weekly exercise. Anyone might think that the longer the time, the less likely you are to develop a major disease, but the reality is that above this time the benefits seem to stagnate. This shows that maximum efficiency is achieved in that limited period of time, demystifying the need to spend endless days in the weight room to obtain many more protective advantages at the metabolic level that allow us to extend our life a little more or make it of a higher quality. You have to combine it. Although strength training shines in this study, abandoning cardiovascular exercises would be a profound mistake. Here the research group itself pointed out that combining strength exercises with aerobic activity offered the best possible results, since this duality confirms that a hybrid approach dramatically maximizes long-term survival benefits. It’s backed up. In the past there were reviews that explored the relationship between training and mortality, this being one more that gives it much more strength so that it ultimately continues to be recommended for consultation to anyone, regardless of age. Because exercise here does not understand age, and strength exercise can be for the youngest, but also for the elderly who need to preserve their muscle to have a better quality of life in their last years of life. Images | Anastase Maragos In Xataka | In the fever to train strength, the gym has faced competition: more and more people train on the street

The world is running out of data to continue training AI. China has an ace up its sleeve

by

The models of artificial intelligence (AI) have a problem that more powerful chips cannot solve: they are running out of data. Epoch AI, a nonprofit research organization specializing in scaling AI models, warns with 80% certainty that the high-quality text available on the Internet will be exhausted sometime between 2026 and 2032. The reason is very simple: AI laboratories have been extracting everything the web has to offer for many years, and current models already train on data sets that approach the theoretical limit of the information available. When that gold mine empties, data volume scaling will stop working. And if this scenario occurs, AI development will most likely slow down. We still do not know what strategy US companies are developing to solve this problem, but we already know what is China preparing. His biggest rival. In fact, Xi Jinping’s government has decided that this shortage is an opportunity. This week the China National Data Administration published a draft outlining its action plan with a clear objective: to build an ecosystem of validated data by 2028 that will fuel the next generation of AI models. China’s bet is already on the table The document prepared by the National Data Administration identifies which specific sectors are priority objectives for information generation and certification. Some of them are scientific research, manufacturing, agriculture, energy, transportation, finance, healthcare, education and e-commerce. However, his plan does not stop at traditional sectors. China has a structural advantage that no Western laboratory can easily replicate And it also plans to cover cutting-edge fields with quality data, such as AI applied to robots, autonomous driving, low-altitude aviation or biomanufacturing. These are, precisely, domains whose data is not on the internet because they come from sensors, actuators and physical environments. Achieving them requires having industrial infrastructure, and in this scenario China has a structural advantage that no Western laboratory can easily replicate. However, this is not all. The document prepared by the National Data Administration explicitly encourages the expansion of the supply of text, code, images, audio and video necessary to train systems capable of complex reasoning, agentic behavior and control of intelligent robots. In fact, it’s an almost exact description of what the industry calls next-generation models. They are not just systems capable of answering questions; They will also be able to plan, act and operate in the physical world. The availability of high-quality multimodal data, especially that coming from real industrial environments, is today one of the least discussed and most determining bottlenecks in the AI ​​career. In a scenario where access to cutting-edge chips is restricted by US export controlsdata becomes a competitive advantage. If China can’t win the hardware race, it can try to win the fuel race that that hardware needs to be truly useful. Image | Daoducquan More information | SCMP In Xataka | The condemnation that afflicts China: after decades of manufacturing a competitive desktop processor, it is six years behind

secret training for war in Ukraine

by

The scene took place a few months ago. Ukrainian soldiers surprised British instructors when they discovered that many NATO armies still They did not use anti-drone networks on a regular basis. After several years of war, Ukraine was beginning to teach the West how to survive on a front dominated by drones. Much more than drones. For much of the Ukraine war, the relationship between China and Russia has been interpreted primarily in terms economic and technological. Beijing bought Russian oil and gas while Chinese companies appeared linked to the supply of electronic components, drones and machinery useful for the Russian military industry. However, the revelations he has had access Reuters on the secret training of Russian military in Chinese facilities point to something much deeper: China would not be limiting itself to indirectly supporting the Russian war economy, but rather participating in the tactical and doctrinal training of soldiers who then return directly to the Ukrainian front. This enormously changes the dimension of the relationship between both countries. War as a military classroom. According to the documents and sources European intelligence agencies, some 200 Russian soldiers were trained discreetly in China at the end of 2025 under an agreement signed between senior commanders of both countries. He program included training in FPV drones, electronic warfare, army aviation, mechanized infantry and demining. Some sessions took place in military centers in Beijing, Nanjing, Zhengzhou or Shijiazhuang. What is important is not only the relatively small number of soldiers, but the profile of many of them: instructors and commanders capable of relaying that knowledge to whole units once back in Ukraine. In other words, China would not simply be sending technology, but helping to perfect the way Russia fights modern war. China learns while Russia fights. It just so happens that the relationship also greatly benefits Beijing. The People’s Liberation Army has not fought a major war in decades and the Russian invasion of Ukraine has become the largest military laboratory real of the planet. Russia brings direct combat experience in drones, trenches, electronic warfare and mass attrition. China provides industrial capacity, advanced simulators, technological production and training methods increasingly sophisticated. The exchange is extremely valuable for both. Moscow gains access to technology and training difficult to obtain under Western sanctions, while Beijing can observe how modern weapons, tactics and doctrines really work without being officially involved in the conflict. Silent revolution. The heart of all this cooperation revolves around drones. Ukraine has completely transformed the way it fights using cheap FPV capable of destroying armored vehicles, fortified positions and even helicopters. Russia had to quickly adapt to that reality and now appears to be turning to China to further professionalize part of that ecosystem. The documents describe simulator training flight, coordinated use of drones with mortars, electronic warfare against enemy drones and physical interception systems through networks. All of this reflects the extent to which modern warfare is ceasing to depend exclusively on large traditional platforms to increasingly focus on cheap, massive and very difficult to neutralize systems. Europe’s concern. For the European agencies, what is truly disturbing is that part of those soldiers trained in China already they would have participated later in combat operations in occupied Crimea and Zaporizhzhia. This means that the knowledge acquired in Chinese facilities ends up being applied directly on the European battlefield. Beijing, for its part, continues to publicly defend a neutral position and continues to present itself as a possible peace mediator, but this type of cooperation seriously erodes that image. In the eyes of many Western governments, China would be entering a much more sensitive gray area: not officially sending its own troops or weapons, but contributing to improving Russian operational capacity in an active war against Ukraine. Increasingly military alliance. The revelation It also confirms the extent to which the “limitless” partnership announced by Xi Jinping and Vladimir Putin before the 2022 invasion has evolved far beyond simple joint exercises or diplomatic statements. China and Russia no longer seem to limit themselves to coordinating political positions vis-à-vis the West, they are beginning to share knowledge combat practices, training and doctrine. The most significant detail may be precisely the secrecy of the agreements: prohibition of media coverage, restrictions on information to third parties and programs developed discreetly away from the international spotlight. All this suggests that both countries perfectly understand the political sensitivity of a cooperation that, although still indirect, gets closer and closer to China to the real workings of the war in Ukraine. Image | Vitaly V. Kuzmin In Xataka | The closure of the Strait of Hormuz chokes the Chinese economy. Its only energy solution is a historic pact with Putin In Xataka | While everyone was looking at Hormuz, Russia has found a much more important route to supply drones to Iran

Schwarzenegger continues training every day at 78 years old and the fascinating thing is that he is right

by

At 78 years old, the seven-time Mr. Olympia and the most famous cyborg in the history of cinema continues to faithfully attend his appointment at the gym, as Schwarzenegger acknowledges in an entry on his personal blogwhere he explains that even on days when he has less energy he goes to exercise, something that sums up pretty well with the phrase: “No matter the pain, no matter the weight, every day I achieve a victory.” It’s a reality. His mentality, forged in the golden age of bodybuilding, might seem like the eccentricity of a Hollywood star who refuses to age. However, behind Schwarzenegger’s weights and pulley machines lies one of the most robust physiological truths of modern medicine: strength training starting at age 70 and 80 is not an aesthetic whim, It is a medical necessity. Falling into the stereotype that sport is for young people who want to show off having a good body on the beach is real nonsense, because playing sport literally becomes the next prescription for all the benefits it entails. And logically being more or less old here does not at all condition entry to a gym, as science recognizes us. Inactivity weighs more. To reach old age in better health, you don’t have to invest a lot of money in super expensive supplements or creams to keep your skin firm. And there is a widespread belief that the loss of functional capacity and weakness are inevitable consequences of getting older, but here the National Institute on Aging from the United States is blunt about it: in the vast majority of cases, physical inactivity weighs much more than biological age itself in this deterioration. One of the great silent enemies of middle age is sarcopenia, which is the gradual loss of muscle mass and strength, and which literally correlates with both quality and life expectancy. And to correct it, the only treatment we have at hand is to do strength training adapted to each person profile, directly improving our ability to perform such everyday tasks as getting up from a chair, walking to the supermarket or carrying shopping bags. It’s never too late. Another of the great myths that we also have on the table is that you are too old to start exercising, but here the scientific literature reminds us that it is false, since starting late is still (very much) worth it. Here, a study that grouped 121 randomized trials with 6,700 participants showed that progressive strength training improves muscle strength and functional capacity in older people. This resulted in an improvement in autonomy by significantly improving their walking speed and their ability to climb stairs autonomously. Besides this, a recent systematic review of the Polytechnic University of Madrid on training in older people reported that traditional strength training can achieve improvements in knee extension strength of up to 46%. That percentage, in clinical practice, is the difference between needing a walker or walking on your own. Beyond the muscle. Strength work is also important for bone formation and combating osteoporosis by reducing several cardiometabolic risk factors. Besides, the Heart Foundation points out that strength work, added to balance and mobility training, is vital to protect against falls. It is not something minor, since in people over 70 years of age a fall is not an accident, but rather it is one of the most serious clinical problems that can drastically reduce life expectancy. Added to all this is that exercise has been shown to help regulate sleep, improve mood (reducing stress and anxiety) and even protect cognitive function. Adapted. There is no need to try to emulate Schwarzenegger’s youth records to obtain these benefits, but current medical guidelines agree on a minimum effective dose that is very affordable for almost everyone. In this case, for a person over 70 years old, a reasonable guideline supported by evidence is to train strength at least 2 days per week. But here you should always adapt the exercise to each person and start with a gentle exercise and gradually increase it. Although you don’t just have to be with the dumbbell in your hand, it should be combined with balance routines, joint mobility and some aerobic work. Images | Wikipedia Victor Freitas In Xataka | We have been debating for years whether it is better to go to the gym in the morning or in the afternoon. Physiology finally has the answer

Pokémon Go brought millions of players to the streets. Millions of players who were actually training an AI

by

In 2016 it came to the mobile market Pokémon Goa spinoff of the popular entertainment franchise with a very interesting premise: capture Pokémon in your city using your cell phone’s GPS. The game caught on very quickly and became a phenomenon. It’s been almost 10 years since that and Niantic, its developer, has taken advantage of all the data that millions of players have been giving them to guide delivery robots through the cities. Your first client: Coco Robotics. The business that no one saw coming. The amount of information that can be obtained from Pokémon Go is truly impressive, since millions of people have voluntarily traveled the world with their mobile phones in order to capture (digitally) this type of creatures. And each game leaves an invisible trace, since there are millions of photos of buildings, squares and streets labeled with very precise coordinates that would not have been possible without the information provided by its users when playing. Five hundred million people installed the app in its first 60 days, according to Brian McClendonCTO of Niantic Spatial. Eight years later, the game still has more than 100 million players in 2024, according to data from Scopely, the company that acquired Pokémon Go from Niantic that same year. The problem that GPS does not solve. GPS devices become a bit silly when they have to operate on sidewalks and much of the urban fabric that does not correspond to the road. Signals bounce between skyscrapers, tunnels and viaducts and the margin of error can be up to 50 meters, enough to place a robot on the wrong sidewalk or the next street. “The urban canyon is the worst place in the world for GPS,” affirms McClendon. Coco Robotics, a startup that operates nearly 1,000 delivery robots in cities such as Los Angeles, Chicago, Miami and Helsinki, knows this well, as its devices operate precisely in those dense areas where the signal is never reliable. This is where Niantic Spatial comes in. In May 2024, Niantic separated its spatial and artificial intelligence division. created Niantic Spatial as an independent company. Its core product is a visual positioning system (VPS) trained with 30 billion urban images, capable of placing a device on the map with a precision of a few centimeters from a handful of photos of the environment. The key is that these images come from millions of points of interest in Pokémon Go and Login (the company’s pre-Pokémon Go AR game, released in 2013). In such popular games, players have for years been directed to photograph the same place from different angles, at different times and in different weather conditions. “We had over a million locations around the world where we can locate you to the nearest centimeter and, more importantly, know where you are looking,” explains McClendon. What this changes for robots. Coco Robotics has been the first partner to adopt this technology. Its robots, equipped with four cameras, will combine conventional GPS with Niantic Spatial’s VPS to position itself more accurately, especially in pickup areas in front of restaurants and in delivery to the customer’s door. According to Zach Rash, CEO of Coco, the goal is meet delivery times promised and not depend on margins of error that in practice mean arriving late or to the wrong place. The model already solves one of the most practical challenges of urban robotics: performing well where conventional systems fall short. Beyond the distribution. John Hanke, CEO of Niantic Spatial, talks about what he calls a living map: a hyper-updated simulation of the real world that updates as robots move through it and provide new data. The idea is not only that the maps are more accurate, but that they are designed for machines, not people. This involves adding descriptions of each element of the environment, its properties, its context. “This era is about building useful descriptions of the world for machines to understand,” says Hanke. In that sense, Niantic Spatial differs from other bets on world models, such as those of Google DeepMind or World Labswhich focus on generating virtual environments. Niantic Spatial wants to replicate the real world as it is. In Xataka | OpenClaw changed the rules of the AI ​​race. Technology companies already have their answer: copy it

If the controversy is that AI steals works in its training, the European Union has the solution: license them

by

A few weeks ago the Washington Post published this image of the “Panama Project”: It is a warehouse with hundreds of thousands of books waiting their turn to be scanned and destroyed in the process. It is part of an internal program Anthropic to train its AI and the result of tens of millions of dollars in purchases to digitize all those works without permission from their authors. They are not the only ones who “they borrow” copyrighted content to train their artificial intelligences and the European Union is clear about something: stop stealing protected content and properly license works to train AI. And AI companies defend themselves by saying that no one is going to think about small companies. Europe is clear: if you want to train AI, pay the author It is curious how the entertainment industry and the regulation of countries shook hands at the beginning of the 2000s with those ads of “you wouldn’t steal a purse. You wouldn’t steal a car. Don’t steal a movie.” They portrayed copying a CD or downloading a movie as if you were breaking into the Pentagon’s systems. Years later, that same industry turns a deaf ear given what big technology companies are doing to train AI. The Washington Post document states that others such as Meta, Google and OpenAI They had also participated in the race to obtain data in bulk for your models. There are kicking examples, like the 81.7 TB of copyrighted books that you have downloaded Meta or that OpenAI will use animation from all the studios to train its AI (earning reproaches by Ghigli and more Japanese studies and complaining that Deepseek has looted ChatGPT). Given the context, it is time to say that the European Parliament has grown tired of this and has one of the things he is best at: legislating. In this case, it makes perfect sense for Europe to take this measure, and the agency issued a report non-binding law that urges the European Commission to develop rules that set minimum standards for these AI companies. “Generative AI should not operate outside the rule of law” Basically, if they use protected content for their training, they must license it and also compensate the authors. with the title “Protecting creative work with copyright in the age of AI”the European Parliament demands a series of measures apart from licensing the works. They are the following: Calls for the transparent and remunerated use of protected content to train generative AI. AI vendors are expected to recognize and pay for the copyrighted work they used to train their systems. Measures so that owners of works with rights can exclude their protected work from training. The reason that they argue MEPs is that “generative AI should not operate outside the rule of law. If copyrighted works are used to train artificial intelligence systems, creators have the right to transparency, legal certainty and fair compensation.” The European Group of Societies of Authors and Composers, or GESAC, points in the same direction. In statements to EuronewsAdriana Moscoso del Prado, general manager of GESAC; assures that “this vote adds to the growing recognition at the EU level of what is at stake. Innovation, equity and cultural sovereignty must go hand in hand.” AI companies fight back From the CCIA, the Computer and Communications Industry Association, it was noted that this is not a measure to protect artists, but rather “a compliance tax.” That is, something that must be fulfilled no matter what and that goes against progress. The group argued that such a measure would not go against large companies, but against small ones. They say that many will have difficulty negotiating complex licensing agreements with major publishers, “holding back Europe’s digital competitiveness on the global stage” and stating that what they would need to do is improve existing laws in the European Union, including the AI ​​Law and the Copyright Directive. In any case, there is nothing on the table at the moment. As we say, it is a self-initiative report by Parliament and is not binding. The Commission can now consider whether to do so or not, but it makes one thing clear: Parliament’s position on any future AI measures by the Commission. The problem is that generative AI has already plundered millions of copyrighted works on which it can build its next interactions. The software has tons of information to pivot on and can evolve in other areas, like stopping hallucinating, for example. And it is another example of the two speeds of this matter: the technological ones taking the first steps and the legislators behind them seeing what can be done when the act they want to legislate on was already carried out years ago. Images | Washington Post, Anti-Piracy Campaign (edited) In Xataka | The AI ​​industry is only sustainable by violating copyright laws. So he’s trying to eradicate them

Tencent has a significant stake in US military training tools. Trump is going to stand up to it

by

The Trump administration is debating if it forces the Chinese giant Tencent to get rid of its stakes in the largest Western video game companies. At stake are Riot Games, Epic Games and Supercell (more than a billion players) and the Unreal Engine, used in military simulations. The ghost of TikTok returns, but this time the affected market is different. Why Tencent. Tencent is not only the largest video game company in the world. It is also the largest silent shareholder in the Western industry: it owns 100% of Riot Games, 28% of Epic Games and majority control of Supercell, the Finnish company behind ‘Clash of Clans’. To this we must add participations in Larian, Remedy, Ubisoft and Discord, among dozens of other studios. For years, that capital has flowed to the West: the studios needed investment, Tencent had liquidity, and no one was looking for trouble. The White House sniffs. Washington, however, he has had doubts for years. The Committee on Foreign Investment in the United States (CFIUS) began to review these investments during Trump’s first termand the case became one of the longest in the history of the organization, going through two administrations without reaching a clear resolution. What worries the White House is that video game platforms collect financial information, personal data and chat logs from hundreds of millions of users, many of them Americans. These databases are candy for any intelligence agency. The Epic case. The Unreal Engine adds an extra issue in which the White House has a special interest. The engine not only gives life to video games like ‘Fortnite’; It is also used by defense contractors and the US military itself for military simulation and training. In fact, the country’s Armed Forces have worked directly with Epic for years on that development. That Tencent is a shareholder in the company that builds this technology is what turns this issue into a national security problem. So much so that in January 2025, the Pentagon formally classified Tencent as a company linked to the Chinese military. Tencent rejected that classification, but the Pentagon did not withdraw it. There are problems. During the Biden administration, the issue was entrenched by an internal disagreement that no one knew how to resolve: Deputy Attorney General Lisa Monaco defended forced disinvestment, but the Treasury Department preferred to keep investments under data segregation protocols. Without consensus, the case was frozen. The cabinet meeting scheduled for March 4 was postponed due to scheduling conflicts. That same day, Tencent shares fell 1.72%. Parallels with TikTok. There are similaritiesbut also differences. With ByteDance, the US forced the creation of a new entity with 80% in the hands of US investors, as a condition of operating there. But the problem with Tencent is that it does not operate on American soil, but rather is a shareholder in companies already established there. Getting rid of these stakes is not the same as closing an app, it is more a restructuring of private capital. The consequences in the case of Tencent would go beyond Riot and Epic: the Chinese company has been the main injector of capital into studios for a decade, and a forced disinvestment would change the financing conditions of the entire sector, favoring large publishers. When will there be a solution? The decision has an undeclared but known deadline: Trump travels to China in April to meet with Xi Jinping. Forcing Tencent to sell would send a message of maximum pressure before sitting down to negotiate. In any case, neither the US Treasury, nor Tencent, nor Epic nor Riot have made public statements. Silence, in this type of situation, is louder than if they were discussing it loudly. In Xataka – China has made a drastic decision: prioritize ‘its’ technology, even if it is worse

AI consumes obscene amounts of energy. Sam Altman compares it to the cost of “training” humans

by

OpenAI CEO Sam Altman participated in an event organized by The Indian Express. During the interview made some striking statements, but the greatest of all of them was the one he dedicated to talking about what it costs to train an AI model. In fact, he complained about how many of ChatGPT’s energy consumption discussions they are unfair. Training humans also consumes a lot. The interviewer asked Altman about ChatGPT’s energy consumption and Sam Altman took a few seconds to answer the question, and then made a peculiar comparison (my bold): One of the things that is always unfair in this comparison is that it talks about how much energy it takes to train an AI model compared to what it costs a human to perform an inference query. But it also takes a lot of energy to train a human. It takes about 20 years of life and all the food you eat during that time before you become intelligent. And not only that, it took the widespread evolution of the hundred billion people who have lived and learned not to be eaten by predators and to understand science and so on to create you. The fair comparison is if you ask ChatGPT, how much energy does it take once their model is trained to answer that question compared to a human? And AI has probably already caught up in terms of energy efficiency if we measure it that way. A previous Epoch AI study corroborates that energy consumption during inference (when we actually use ChatGPT, for example) is low. Source: Epoch AI. Training is one thing, inference another.. The answer may be controversial, but to a certain extent it is logical: learning, both in the case of humans and AI, takes time and consumes many resources, but that cost is one thing and the cost of inference, of “applying that training”, is another. Once we have learned, it is not too difficult to answer things. This is what Altman is trying to point out here, who recognizes that AI does indeed consume a lot of energy in training, but that it has then become very efficient in the inference phase, when we actually use ChatGPT. The problem is that although Altman has already spoken that in inference consumption is minimal, does not provide evidence of this. The water problem is no longer a problem. He also spoke about the controversial water consumption that was theoretically carried out in large AI data centers. Although he acknowledged that this was a problem when “we used to use evaporative cooling in data centers.” Now, however, “we don’t do that,” he recalled, and made it clear that those accusations that “ChatGPT uses 17 gallons per query, or whatever” is totally false, “totally crazy, it has no connection with reality.” But again, there is still no official data from AI companies in this section. How much does AI really consume? The truth is that at this point we still do not have really clear data on how much the AI ​​consumes both in the training phase and in the inference phase. There are those who have investigated energy and water consumption and have made a mistake. wildly exaggerating the databut for example in the US, where a large number of data centers are concentrated, there is no legislation that forces transparency with those figures. Increasingly more efficient models and data centers. One of the most interesting studies was the one made by Epoch AI in February 2025, and at that time it was also concluded that AI did not actually consume as much as it was said to consume. In fact, it consumed relatively little and the models have only improved in efficiency. Chips and cooling systems have also improved, and although data centers have certainly require enormous amounts of energywe continue blindly in this section. In Xataka | Spain has a plan to capture more data centers than anyone else: “shield” them from energy costs

The best science comedian does not have any scientific training. And that’s the key to your success.

by

Tom Gauld is one of the most accessible and yet peculiar cartoonists of today. His vignettes are a mixture of a wink for the initiated and simple, white humor.which often makes his cartoons a mix of “everyone can understand them” and “if you’re interested in science and literature, sure.” A real rarity in these times when you have to show up at franchise fan clubs with a very clear identification and resume. Because Gauld may talk about quantum physics, multiverses and the secrets of the cosmos, but he doesn’t leave anyone out either, all thanks to deceptively simple, but highly expressive graphics. Able to make an Escherian architectural nonsense believable or to perfectly portray the interior of an impossible dimension with just a couple of lines, Gauld reduces the complex to a couple of gentle strokes, and hence his popularity on the internet and in media of indisputable prestige such as ‘The Guardian’where he makes literary jokes, or ‘New Scientist‘, where it focuses more on science and technology. It is precisely a compilation of jokes of this last type, ‘Physics for cats’, which Salamandra is now publishing. Thanks to this brand new volume we have had the opportunity to speak with him and have him explain his creative processes and his career as a scientific comedian… who does not have much knowledge of science. We started, of course, by asking him how his collaboration with ‘New Scientist’ began and what impact it has had on the way he approaches scientific topics in his comics. It tells us that we have to go back very far in time. “My grandfather was a scientist, a marine biologist, and he always read the ‘New Scientist’. So when he went home, the magazine was always there, and when he finished reading the magazine, he would give it to my father, who was also interested in science. When I was little, I would look at the pictures and diagrams and, from time to time, I would read a little bit of the text.” And from there, a few years later and now a professional cartoonist, he began to collaborate with them. Gauld states that a magazine of this type is a splendid workplace for an illustrator: “Some concepts about reality or other universes cannot be photographed, so in These types of magazines have a good tradition of using illustrationsand in fact most of its covers are illustrations rather than photographs. Then, I don’t remember exactly why, I thought it was strange that they didn’t have a comic strip in the magazine.” He proposed it a decade ago and it was accepted, but, he says, “I got a little scared because I stopped studying science when I was about 16, so I’m not an expert at all.” How to draw science It is obvious that this approach to science from a non-scientist perspective will entail difficulties. But contrary to what it might seem, “the really difficult thing with vignettes is not getting the scientific details right.” His process is: “I read the magazine, I follow scientists on social media, I listen to podcasts and radio shows about science, and anything that I think could make a joke I write down in my notebook.” And his approach is clear: “I’m giving my own light-hearted, fun take on something that’s quite serious and thoughtful. I try to do it without being derogatory, like when you make fun of a friend you respect.” Which inevitably brings us to the next question: how do you balance scientific precision with the artistic freedom to create such abstract concepts? And in fact, here the lack of scientific training is revealed as an advantage: “When creating the strips, the fact that I have no scientific training, that I am an ordinary person, not a professional, perhaps helps me judge the level of knowledge at which the jokes should be.” And he adds: “I never want to make a cartoon that makes people feel stupid.which makes one think that a doctorate is needed to understand it”. What happens then when he stumbles upon concepts that even he can’t understand? “When some real science is mentioned in the cartoon, I like to get it right, so I do some research on the Internet or ask someone at New Scientist to check my formulas or whatever. Or I do it so badly that it’s obvious I’m not trying to get it right. In fact, last night an astrophysicist mentioned that one of the formulas in the background of one of my strips was correct and that he liked it, which I was very happy about.” When we ask him if there are any scientific ideas or theories related to physics that he finds especially inspiring, he tells us that two come to mind. “One that I think I keep coming back to in the cartoons is, and I guess this is more of a philosophical question than a physical one: What is reality? That and the idea of ​​many worlds. The other is quantum theory, which I still don’t understand. I’ve made some jokes about it and I’m proud of them, but I think they could be improved if I ever managed to understand all of quantum theory. Which may never happen, but I keep trying.” And here we enter into a personal question, but we couldn’t help but ask him: does Tom Gauld like Gary Larson’s humor? (Larson, for those who don’t know, is the creator of ‘The Far Side’, absolute master of comics with background geeka mix of surreal humor and deep knowledge of biology and science absolutely unmatched). “I’ve mentioned Gary Larson as an influence in almost every interview I’ve done today,” he confesses, “so I’m glad you brought it up.” Typical Gary Larson: “‘Hey! What is this, Higgins? Physics equations?… Do you like your job as a cartoonist, Higgins?” And he adds: “The cartoons from ‘The Far Side’ appeared in my local newspaper when I was a teenager and I have … Read more

The industry became obsessed with training AI models, while Google prepared its masterstroke: inference chips

by

In recent years, what was truly relevant was training AI models to make them better. Now that they have matured and training it no longer scales as noticeablywhat matters most is inference: that when we use AI chatbots they work quickly and efficiently. Google realized this change in focus, and has chips precisely prepared for it. Ironwood. This is the name of the new chips from Google’s famous family of Tensor Processing Units (TPUs). The company, which began developing them in 2015 and launched the first ones in 2018now obtains especially interesting fruits from all that effort: some really promising chips not for training AI models, but for us to use them faster and more efficiently than ever. Inference, inference, inference. These “TPUv7” will be available in the coming weeks and can be used to train AI models, but they are especially aimed at “serving” these models to users so that they can use them. It is the other big leg of AI chips, the really visible one: one thing is to train the models and quite another to “execute” them so that they respond to user requests. Efficiency and power by flag. The advance in the performance of these AI chips is enormous, at least according to Google. The company claims that Ironwood offers four times the performance of the previous generation in both training and inference, and is “the most powerful and energy-efficient custom silicon to date.” Google has already reached an agreement with Anthropic so that the latter has access up to one million TPUs to run Claude and serve it to its users. Google’s AI supercomputersand. These chips are the key components of the so-called AI Hypercomputer, an integrated supercomputing system that according to Google allows customers to reduce IT costs by 28% and a ROI of 353% in three years. Or what is the same: they promise that if you use these chips, the return on investment will be multiplied by more than four in that period. Almost 10,000 interconnected chips. The new Ironwoods are also equipped with the ability to be part of joining forces in a big way. It is possible to combine up to 9,216 of them in a single node or pod, which theoretically makes the bottlenecks of the most demanding models disappear. The size of this type of cluster is enormous, and allows for up to 1.77 Petabytes of shared HBM memory while these chips communicate with a bandwidth of 9.6 Tbps thanks to the so-called Inter-Chip Interconnect (ICI). More FLOPS than anyone. The company also claims that an “Ironwood pod” (a cluster with those 9,216 Ironwood TPUs) offers 118x more ExaFLOPS FP8 than its best competitor. FLOPS measure how many floating-point math operations these chips can solve per second, ensuring that basically any AI workload is going to run in record times. NVIDIA has more and more competition (and that’s a good thing). Google chips are a demonstration of the clear vocation of companies to avoid too many dependencies on third parties. Google has all the ingredients to do it, and its TPUv7 is proof of this. It’s not the only oneand many other AI companies have long sought to create their own chips. NVIDIA’s dominance remains clearbut the company has a small problem. In inference CUDA is no longer so vital. Once the AI ​​model has been trained, inference operates under different game rules than training. CUDA support remains a relevant factorbut its importance in inference is much less. Inference focuses on obtaining the fastest possible answer. Here the models are “compiled” and can run optimally on the target hardware. This may cause NVIDIA to lose relevance to alternatives like Google. In Xataka | When you’re OpenAI and you can’t buy enough GPUs, the solution is obvious: make your own

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.