data

While we debate how much electricity AI needs, China already has a data center that runs on clean energy

by

The first data center Artificial Intelligence Powered by China’s Wind Power. This is an innovative facility in the Ningxia Hui Autonomous Region, in the northwestern part of the country, and has a renewable energy supply that allows it to operate using 100% clean electricity. This means everything a milestone for the Asian country and also for the global industry if we take into account the debate that sparks the enormous environmental impact of these infrastructures. Less emissions into the atmosphere. According to what the media shares South China Morning Postthe IA data center’s renewable energy plant generates 4.3 billion kilowatt-hours of electricity per year, reducing carbon emissions into the atmosphere by 3.65 million tons. Furthermore, the installation reaches a 1.15 in the energy use efficiency indicator (PUE), which measures the entire power of a facility and the energy expenditure of data centers. To put this in context, the industry’s average PUE has been 1.56 over the last 5 years, according to the Global Data Center Survey 2025 by Uptime Institute. A score of 1.0 would be a perfect PUE, according to this organization. Data in the East, Computing in the West. The implementation of this AI data center is part of a broader strategic plan that the Asian country is carrying out to reinforce the infrastructure that supports the development of AI using renewable energy. With the name in English Eastern Data, Western Computing (EDWC), this national megaproject seeks to move data processing from the more developed eastern areas of China to the rural western regions, which have much better and abundant environmental conditions for renewable energy sources. This would be the case of the Ningxia Hui Autonomous Region. In this way, the Asian giant plans to face the increase in energy demand in the coming yearswhich may be double in 2030 according to some forecasts. Voracious consumption. The AI data centersdespite being very similar to traditional data centers, work with a greater processing demand, which makes them true energy guzzlers. The ICT consulting company Gartner predicts that the electricity consumption of this type of locations will reach 565 TWh this 2026. Regarding its impact, a recent study carried out by Allianz reveals that the emissions associated with this type of centers reached 286 megatons of carbon dioxide in 2025 and their carbon footprint has been 57% larger than what the IEA had originally calculated. Among the resources that have been used the most, electricity consumption is at the top, responsible for 76% of that footprint. For this reason, the report suggests that moving towards a greener AI will depend, above all, on transforming the energy systems that feed it and betting on renewable resources such as wind energywhich has hardly any water or carbon footprint. And here, what are we doing? The “green” outpost What China is taking is of enormous interest for the rest of the world, especially for our continent if we take into account the decarbonization objectives that come from Brussels. The European Union has already contemplated this type of data centers in some of its general energy regulations, such as the Energy Efficiency Directive (EU) 2023/1791which in its article 12 establishes that those centers with more than 500 kW will be obliged to make their energy consumption public. Other standards, such as Delegated Regulation (EU) 2024/1364seek to unify a regime of common assessment for the more than 3,000 installations of this type that exists in the EU. Although there is no shortage of those who see in this regulatory will a reason why Europe could be left behind In the technological race, the truth is that the energy efficiency of data centers, and especially in AI centers, can make a difference. That’s how he thinks Hongpeng LeiHead of the Mitigation Division of the United Nations Environment Program (UNEP), for whom betting on more sustainable centers is not an obstacle, but “the path to a resilient and competitive digital infrastructure”. Spain, for its part, is one of the European leaders in renewable energy with more than 55% of the electricity produced with this origin in 2025. This obvious advantage gives it a great potential to build a green-powered next-generation data infrastructure. However, it is worth asking whether the Spanish network will be able to keep pace with the installation of these centers and, consequently, satisfy what is clearly going to be a huge energy demand. For now, The project that Microsoft plans to build in Aragon will consume more energy than the entire autonomous communityan example of the energy challenge that accompanies the development of AI. Image | Paul Hanaoka Xataka | “Only behind the United States”: China has achieved a until recently chimerical position in intelligent computing

The most valuable company in the world has no interest in investing in the data center fever

by

Big Tech is investing absolute fortunes in data centers. Almost all of them seem to want to be prepared to meet the colossal demand for computing capacity for AI tasks. And yet, there is a curious and surprising exception. Money, money and more money. The big technology companies have put all the meat on the grill with billion-dollar investment plans in data centers throughout 2025. Amazon, Google, Meta and Microsoft jointly invested more than $416 billion in capex during 2025, 66% more than the $251 billion in 2024. Google reached $91.5 billion, marking a new all-time high; Meta’s investment reached 72.2 billion dollars and Microsoft reached 118 billion. By 2026, the climb continues: Amazon, Microsoft, Alphabet and Meta are jointly committing about 725,000 million dollars in AI infrastructure in 2026, 77% more than in 2025, and analysts already project that the figure will exceed one billion in 2027. To give us a slight idea of ​​what that figure means, it is higher than Sweden GDP in 2025 (662,000 million), that of Israel (610,000 million) and that of Singapore (574,000 million). Apple. However, at Apple things are very different, and its Capex for 2025 remained the same. almost flat compared to 2024 with a total investment of 12,715 million dollars. It is a notable figure, of course, but it is far from the 20 billion dollars that Apple plans to invest in 2026. And she is the one with the most money. It is also interesting to compare the investment that these companies will make with respect to their market capitalization. Amazon is the biggest bet here: its capex is equivalent to approximately 7.63% of its current market capitalization, followed by Microsoft, with 6.67%. Further behind appears Alphabet, with around 4.49%, and Meta, with around 3.04%; In contrast, Apple, which remains the most valuable company in the world, has a capex of only 0.29% of its market value. Wall Street rules. The decisions of these companies seem to be greatly influenced by the behavior of the stock markets. The presentation of financial results and short-term forecasts has caused a kind of contagious effect: if we do not invest a fortune in data centers, shareholders are going to punish us a lot, Big Tech seem to have said with those colossal figures. The big beneficiary is NVIDIA. Meanwhile, it is likely that the big beneficiary of these colossal investments will be NVIDIA, which will be the one that will receive a good part of that money if everything continues the current trend. There will undoubtedly be other beneficiaries in this area, but NVIDIA’s Capex is comparatively ridiculous, both with respect to those companies and compared to its current market capitalization ($4.75 billion). Apple is going at a different pace with AI. The Cupertino firm clearly seems to have a very different strategy than its rivals in this segment. Google and Apple subscribed a collaboration for which the next generation of its assistant that will be based on Gemini models and Google cloud technology to power future Apple Intelligence featuresincluding a more personalized Siri. The final deployment came at WWDC 2026: Apple presented “Siri AI”, a remake of the assistant integrated with its Apple Intelligence system developed in collaboration with Google’s Gemini models. The feeling, even internally, is that at Apple they are clear followers and that is not going to change at the moment. A version of this article was published in February 2025. Image | Unsplash (Bangyu Wang) In Xataka | Is Tim Cook the Ballmer of Apple and is Nadella the Jobs of Microsoft?

a data center turned neighborhood life into a nightmare

by

For years, data centers remained out of our sight, although much of what we do on the Internet depends on them. In Dowagiac, a city in southwest Michigan, that infrastructure has become impossible to ignore: several residents say that a facility dedicated to cryptocurrency mining and currently in the process of reconversion It has been emitting a high-pitched hum 24 hours a day since 2024. Now, as the company prepares the leap into computing for artificial intelligence and advanced robotics, the conflict shows the extent to which a digital installation can transform the daily lives of those who live around it. unbearable noise. The story focuses on Louise Avenue, a dead-end street where Lindy Valenzuela and her husband and wife Marjorie and Billy Finn live right across the street from the compound, owned by Alliance Cloud Services LLC, a subsidiary of Hyperscale Data. According to what they told the American network WXYZfor decades on the other side there was an industrial building hidden behind a row of pine trees, integrated into the daily landscape of the neighborhood. Neighbors date the change to 2018, when the trees disappeared and a loading dock was installed. Around 2021, a cryptocurrency mining operation began there and, three years later, the operation of the place began to feel different. inside the houses. Valenzuela described the sound with a very specific comparison: “It sounds like having a vacuum cleaner in your living room (…) when the filter is clogged, so it makes a high-pitched whistle. And they left it on and left.” Its consequences appear in everyday gestures that were previously normal. The Finns say they can no longer open the windows like they used to, while Valenzuela says that staying outside for a few minutes on hot days can end up giving them a headache. From complaints to the courts. The conflict ceased to be solely a neighborhood conflict in March 2026, when Dowagiac passed its first industrial noise ordinance. The standard sets a maximum of 65 dB during the day and 55 dB at night, and the municipal official assured that the city had already sanctioned Hyperscale Data for failing to comply. The company, however, questions both the measurements and the method used. In parallel, Valenzuela appears as a plaintiff in a class action that accuses the operator of negligence and substantially interfering with the use and enjoyment of the homes. The turn towards AI: Hyperscale Data has put figures to the transformation announced for the venue. Its CEO, William Horne, explained during an extraordinary meeting of the municipal council that the company plans to invest $100 million to expand computing for artificial intelligence and advanced robotics, while abandoning cryptocurrency mining. It has also acquired adjacent land that it intends to use as a natural separation strip to reduce the sound that reaches the homes. For now, however, local authorities say they have not received any permit requests for this expansion. The answers. Among the responses offered by Hyperscale Data is the progressive shutdown of mining machinery, which the company hopes to complete over the next three months. The CEO went further and stated that they would buy the homes of those who still could not enjoy them. For those affected, this alternative does not resolve what these homes represent. For example, the Finn house has been in the family for almost 100 years and there they maintain their friendships and a support network that also helps Marjorie, diagnosed with Parkinson’s. The case. What is happening in Dowagiac shows how a computer infrastructure can become very physical when it is installed next to homes and operates continuously. Data centers support services that we use every day. This expansion can provide investment and employment, although it can also increase the demand for electricity and water, the occupation of land and, in certain cases, the noise endured by the environment. Images | Brett Sayles In Xataka | The Chinese AI startup that launched the amazing GLM-5.2 model has just reached another milestone: creating its own data center

Nvidia is willing to guarantee $250 billion for OpenAI to have its own data center, according to WSJ

by

Nvidia is considering becoming the main financial driver of one of the largest AI infrastructure projects to date. And according to has advanced The Wall Street Journal, Jensen Huang’s company is negotiating to offer a guarantee of about $250 billion so that OpenAI can rent a colossal data center campus in Ohio, USA. All, without the agreement being closed yet. What exactly is the agreement about?. Nvidia’s endorsement would not be direct money for OpenAI, but rather a financial guarantee. According to the media, it would help those responsible for ChatGPT can lease a 10-gigawatt project that SoftBank’s energy subsidiary, SB Energy, is building in southern Ohio. With the backing of Nvidia, this project could obtain debt financing on better terms, something especially relevant because OpenAI, being a private company that has not yet made a profit, does not have sufficient credit rating to be able to invest in this type of project on its own. Why is such a guarantee necessary?. Without the backing of a creditworthy partner, it would be much more expensive for SoftBank to borrow the money needed to build such infrastructure. Just like share WSJ, by putting its signature as a guarantee, Nvidia reduces the risk that lenders assume and makes the operation cheaper. In the jargon of the sector this is called “credit wrapper”, and it is not the first time it has happened. In fact, Google has endorsed Anthropic data centers before, in part to boost sales of its own AI chips, according to collect the same medium. Figures that give vertigo. The entire project, also counting the chips that will go inside the facilities, could exceed $500 billion, which would make it the largest data center announced to date. The guarantee of 250,000 million would cover the rent and construction debt, but not the semiconductors. For this, Nvidia is separately negotiating another financing for the purchase of chips by OpenAI, which could reach an additional $350 billion, according to share from WSJ. The role of the Government of USA. The project’s electricity is under government control and is financed, in part, thanks to a recent trade agreement with Japan, which committed to invest $33 billion in a natural gas plant in the same area, as remember the WSJ. Secretary of Commerce Howard Lutnick is directly involved in deciding who has access to that energy. The chosen site is, in addition, a former uranium enrichment center already closed, about 80 kilometers south of Columbus, which would allow us to avoid a good part of the problems of permits and opposition from residents around the area that other similar projects in the country suffer. She is not the only one interested. OpenAI has been in talks for weeks to take over that space and it is, according to the WSJ, the candidate that has shown the most interest. But she’s not alone, as Anthropic, Microsoft and Google have also spoken to Lutnick about the same issue in recent weeks. Getting energy power has become the big bottleneck for all companies training large-scale AI models. What does each party gain?. For OpenAI, this agreement would represent a change of model, as it would be its first data center as a direct tenant, a step towards controlling part of its own infrastructure instead of depending exclusively on renting capacity from Microsoft, Amazon and Oracle, as it does now. The company has already raised its computing spending forecast until 2030 to $750 billion, compared to the $600 billion it managed at the beginning of the year. For Nvidia, with a market value close to $5 billion, endorsing these types of projects ensures future demand for its chips for years. The shadow of circular financing. This type of operation, in which large technology companies finance infrastructure that later generates business for themselves, worries a large part of the market. Gary Tan, portfolio manager at Allspring Global Investments, counted to Bloomberg that “although Nvidia’s investments and partnerships bolster confidence in long-term AI builds, investors remain concerned about circular financing.” Nvidia itself has warned in its latest annual report that these types of agreements could reduce its short-term liquidity and increase its exposure to the credit risk of its clients. And now what. Negotiations are still in an early phase and conditions could change or even break before anything is signed, according to quote from Bloomberg. The first phase of the project, with about 800 operational megawatts, would not be ready until 2028. So it seems that we will have to wait to find out if it ends up materializing and how exactly it does so. Cover image | Steve Juvetson, NVIDIA In Xataka | “The world doesn’t realize how much the AI ​​industry depends on scientists in China” Jensen Huang, CEO of Nvidia

create your own data center

by

Z.AI has been the protagonist of big headlines in recent weeks, especially after the launch of its very powerful GLM 5.2 language model. Formerly known as Zhipu, the company has now just finished building a massive 1 gigawatt data center that, according to they count from Bloomberg, it will work exclusively with semiconductors made in China. To put it in context, it would be the most ambitious bet yet by a Chinese technology startup to completely dispense with Nvidia. They can’t buy Nvidia chips. Since January 2025, Z.AI has been on the US Department of Commerce’s entity list, which prevents it from legally purchasing Nvidia chips. The most curious thing of all is that this veto, far from stopping it, has pushed it to build all its future infrastructure relying only on national suppliers, according to they explained from Bloomberg. But it is not at all strange, since it is a movement that fits with Beijing’s strategy of reducing its technological dependence on Washington in the midst of the AI ​​race. In detail. As shared by the media, the data center has already begun to operate partially and is designed to train and deploy the company’s family of GLM models. With a power of 1 gigawatt at full capacity, from Bloomberg they assure that the facility could consume as much electricity as about 750,000 homes need, making it one of the largest data centers ever built by a Chinese artificial intelligence startup. For now, the company has installed at least 10,000 chips from China there, although there is no information on what proportion of the total planned capacity that figure represents. Between the lines. Although the media does not detail which company provides the chips, everything points to Huawei. Z.AI already presented its GLM-5.2 model in June, trained entirely with Huawei Ascend accelerators and without resorting to Nvidia hardware. The model also came out ahead of the rest in rankings of open source models such as Design Arena. The problem is that, according to they claim Bloomberg, Chinese chips like the Ascend still perform worse per watt than Nvidia’s current Blackwell, which means that one gigawatt of Chinese power is not equivalent, in practice, to the same computing power as one gigawatt consumed with Nvidia hardware. Added to this is a manufacturing bottleneck. From Tom’s Hardware they point out that SMIC’s most advanced and stable process (equivalent to a 7-nanometer node) is already operating at over 93% of its capacity, and that the shortage of domestically produced HBM memory limits how many accelerators Huawei, which last year shipped around 812,000 AI chips, can assemble. Bloomberg pointed In addition, citing a previous report, Huawei plans to manufacture about 1.6 million Ascend cores in 2026, enough for about 600,000 chips of its flagship model, the 910C, while Cambricon, another of Nvidia’s major Chinese rivals, has set a goal of half a million accelerators this year. In short: China can build data centers faster than it can make the chips needed to fill them. sweet moment. The Z.AI thing comes at the height of the pulse among Chinese AI startups. Its great rival, Moonshot, had to suspend new subscriptions to its platform last Sunday to prioritize the computing capacity of its current users, after the launch of his amazing Kimi K3 modelaccording to explained the company itself on social networks. Z.AI, for its part, is going through a good financial moment. And it is that according to sources consulted by Bloombergthe company is on track to reach $1 billion in annual recurring revenue after meeting its 2026 sales target this month, and has recently bolstered its capital with an IPO in Hong Kong and a subsequent expansion. The company’s shares have closed with an increase of 37% this Tuesdaycoinciding with the news of the new data center. A more ambitious plan. The Z.AI case is part of a much broader plan by the Chinese Government. And Beijing is preparing an investment of about 2 trillion yuan (about $295 billion) within five years to build a national network of AI data centers, with at least 80% of the technology coming from Chinese suppliers, according to they count from Tom’s Hardware. And in that plan, Z.AI, Moonshot, and the rest of the AI ​​startups that are beginning to deploy their weapons are included. Cover image | Z.AI and İsmail Enes Ayhan In Xataka | ASML sells the most important machines in the chip industry. Now everything indicates that they will be more expensive

Europe forces Google to open Android to ChatGPT and Claude, but the blow is that it will have to give up its search engine data to the competition

by

The European Commission has ordered Google to open two of its greatest competitive advantages– Android’s access to rival AI assistants and the search data that feeds Gemini. They are two binding measures under the Digital Markets Act (DMA)adopted on July 16. Android must be open by July 2027, search data by January 2027, and no resources are stopping those deadlines. In detail. The first measure affects how a rival assistant can live inside Android. Right now, Gemini has privileges that no competitor shares: Keyword activation at the operating system level, not the application level. Access to the home button or navigation gesture as a gateway. Reading content on screen to respond with context. Executing tasks within other applications, such as sending an email or sharing a photo. Priority access to hardware and AI models built into the device. There are 11 points in total. Starting with the next major version of Android, ChatGPT, Claude or any other attendee will be able to claim the same treatment that only they have today Gemini. The second measure is the one that weighs the most for the business: the data of queries, clicks and results with which Google trains a good part of its LLM. From January 2027, rivals such as OpenAI will be able to access an anonymized version, under conditions that Brussels demands are fair and non-discriminatory. Google’s waterline. Between the lines. Google completed a few months ago the transition from Google Assistant to Gemini as Android’s default assistant, just before Brussels opened the file. The order arrived with Gemini already installed as a part of the system and not as just another application. Apple chose the opposite path: it negotiated before launching, proposed a monitored permit system and Brussels rejected it. Siri AI It still hasn’t reached the iPhone in the European Union. Google launched Gemini without asking permission and has discussed the conditions afterwards, with better results. Yes, but. Google, like Apple, alleges security and privacy reasons. Its head of global affairs, Kent Walker, has explained that opening screen reading and the execution of tasks between applications to any external service bypasses the filter that the manufacturers themselves apply today before granting these permissions. And now what. Google will appeal, but the appeal does not give it extra time: from the criteria established in the Apple case, the obligation is fulfilled while it is discussed whether it was legal to impose it. The United Kingdom and Japan, which are preparing their own digital competency frameworks, are presumably taking note of how “fair access” to a mobile is defined here. What Brussels sets about wake words or search data prices will not remain only in Europe. In Xataka | AI has destroyed traditional captchas. Google’s solution requires scanning a QR… and having its services installed Featured image | Xataka

The sea in Benidorm has warmed three degrees in just five years. The data is scary, but science has an explanation

by

The waters of the Mediterranean have been undergoing drastic changes in temperature in recent years and the coast of Alicante has become “ground zero” for this phenomenon. Here a recent report indicates that the sea temperature on the coast of Benidorm has experienced a rise of 3 °C in the last five yearsand although it may end up scary, it has an explanation. The temperature difference. The thermal jump detected in Benidorm is real and, if we look at the records of the Valencian coast, at the beginning of July 2024 the sea surface temperature was around 24 °C. But now in July 2026, the thermometers read between 26 and 28 °C. That is, an anomaly of up to 4 degrees in just a couple of years in specific areas. However, in climatology a background trend is not the same as an extreme event, and we are clear that the rise of 3 °C in five years reflects a marine heat wave, an extreme peak of the situation. As expected. If we look at the full movie, the data from the Copernicus Marine Service confirm that the long-term warming rate of the Mediterranean is 0.4 °C per decade. This means that the “base” temperature should have risen by about 0.12 °C in the last five years, and not 3 °C, so what is happening off the coast of Benidorm is a radical anomalous event fueled by the lack of winds, the anticyclone and high sunshine. We continue to increase. The fact that it is a peak that can end up being recovered and not the base trend does not mean that we should breathe easy, since marine heat waves are like fever: a symptom that the patient is going through a difficult time, and the Mediterranean is the reality that is not going through its best moment. The data of SOCIB in its 2025 report They pointed out that the year was the second warmest year since 1982, with an average sea surface temperature of 21.1 ºC, only surpassed by 2024. But the most revealing data comes from the CLIVAR-Spain program of the AEMET and the IEO-CSIC when they point out that The waters surrounding Spain are warming at a rapid rate between 60% and 200% higher than the global average. The Mediterranean bears the brunt, warming up to three times faster than the average world ocean. The transition. This summer of 2026 is being especially hard, since at the end of June, the AEMET recorded that the Mediterranean reached 26.63 °C on average, being 2.6 °C above normal. And if we go outside the average, in the western basin, the Copernicus satellites have managed to capture astonishing peaks of up to +8 °C above the historical averages (1991-2020) in very localized areas. The importance for Benidorm. A sea at 28°C alters ecosystems, endangers Posidonia meadows, attracts invasive species and, from a meteorological point of view, acts like a can of gasoline. Such a hot sea does not cool the night atmosphere on the coast, aggravating the torrid nights, and accumulates an immense amount of energy that, when colliding with the first gusts of cold air in September, can trigger more violent damage and torrential rains. Images | Emilio Sanchez Hernandez In Xataka | Sea temperature continues to rise and experts are clear: “It is partly favored by the start of a new El Niño phenomenon”

The argument of job creation has been the great asset to build data centers. It would be nice if it were true

by

data centers they consume a lot of electricitymaking raise the electricity bill, they pollute the air and sometimes water too. There are more and more arguments against them, while the arguments in favor for those who live near one of these mastodons are basically reduced to one: they generate many jobs. Sure? What is happening. The United States concentrates a third of all data centers in the world. They count in Futurism that there are states competing to attract big technology companies, offering resources and tax exemptions in exchange for them generating jobs and helping to develop their communities. However, it is increasingly clear that the creation of stable and lasting employment is an unfulfilled promise. Billions for a handful of jobs. In Cedar Rapids, Iowa, there are two projects underway for two data centers from QTS and Google respectively. As reflected in the Cedar Rapids Economic Development Centerto encourage the construction of these infrastructures, the city has offered them a 70% exemption from property taxes for 20 years, in addition to returning 75% of electricity rates (a municipal surcharge that is applied to the bill). Google and QTS are going to spend 1.3 billion on their data centers, but it is estimated that they will save more than 580 million on taxes and municipal fees. And you will say how many jobs are they going to create? Under the contract, they are only required to create 61 permanent jobs. Why is it important. Data centers have become the engine of the American economy, but one thing is the wild investment of big tech to build them and the impact on local employment is quite another. They are enormous infrastructures, which cost a lot of money, but once launched they operate practically on autopilot. The tasks that really need human intervention are very few compared to their magnitude. In Aragon too. This situation is not exclusive to the United States, the same thing is also happening in Aragon. We recently learned the employment figures that Amazon will generate in Villanueva del Gállegowhere they are building six data centers. The company had spoken of up to 29,900 full-time positions linked to the project, but they were counting indirect and induced employment, such as suppliers and associated services. The reality is that between six data centers, they are going to hire only 180 permanent people who will work in shifts so that they can operate 24 hours a day, 365 days a year. Once again, the narrative of mass employment is leaking everywhere. Employment yes, but temporary. To build data centers of that magnitude, many companies from different sectors must move, which indirectly does generate a lot of employment. In the case of Cedar Rapids, the city boasts that the projects will create “thousands of construction and trades jobs,” the problem is what will happen to all of them when the data center comes online. According to a recent report by Turner & Townsend echoed in Futurism, the urgency to build new data centers has caused prices to rise in other projects such as housing construction, in addition to monopolizing specialized labor in construction areas. Image | Xataka with Magnific In Xataka | While most oppose AI data centers, there is one group enthusiastic about them: merchandise thieves

Anthropic claims to have found Claude’s “consciousness”, although the data tells another story

by

Anthropic has published an article that promises to change the way we understand the inner workings of your business models. artificial intelligence (AI), like we explain to you last week. This company claims to have identified a kind of global workspace within Claudewhich he has named J-Spacewhere the model analyzes and manipulates concepts before these become part of the final answer. What is surprising is that, according to Anthropic, this space has not been deliberately programmed: it has emerged as a byproduct of training the model. To analyze it and understand how it works, a technique known as J-Lens which is capable of mapping the internal activations of the model onto words from its own output vocabulary. When asking Claude to solve a multi-step mathematical calculation the visible answer contained only the final result, but within the J-Space each intermediate step was represented separately. Without this technique no part of that process could have been observed from the outside. However, Anthropic itself recognizes that much of what a language model does does not even go through this internal space. Speaking fluently, remembering simple facts or applying basic grammatical rules almost completely dispenses with J-Spaceand when the company blocked its use Claude continued to function normally, although he lost some of his higher-order cognitive functions. Claude “knows” that he is being watched During one of his most striking experiments, Anthropic subjected Claude to injection attempts. prompts during data acquisition processes. He J-Space then showed words like “false”, “injection”, “erroneous”, “prompt“, “fraud” and “poison”, while the visible response of the model completely ignored the attempted manipulation. Interestingly, when pressed to comment on the matter, terms such as “transparency”, “honesty” and “truth” emerged in the internal space. The language used in the article sometimes borders on the idea of ​​an emerging consciousness The Anthropic researchers also observed that Claude seems to recognize when he is being evaluated, which leads him to behave differently when faced with tests than when faced with prompts everyday. In a particularly revealing test, Anthropic removed J-Space the vocabulary associated with that consciousness of being put to the test (terms such as “fake”, “fictitious” or “scenario”) and found that Claude became much more vulnerable to attempts at blackmail or provocation. Be that as it may, the researchers’ report admits important limits: the J-Space is restricted to a single-token vocabulary, so plans or concepts that cannot be expressed with a single word may not appear in the reading, even though the model is calculating them anyway. Anthropic also does not dare to claim that monitoring this space is sufficient to guarantee the alignment of a model. However, we must not overlook that the language used in the article sometimes borders on the idea of ​​an emerging consciousness. Neel Nanda, head of model interpretability at DeepMind, supports this finding as real evidence of a cognitive space within the modelsalthough he clarifies that the practical usefulness of J-Lens remains limited. This achievement, in any case, opens a promising avenue for auditing the honesty of models, although it is still far from being a complete window into a machine’s thinking. Image | Generated by Xataka with ChatGPT using a prompt created with Claude More information | Anthropic In Xataka | While most oppose AI data centers, there is one group enthusiastic about them: merchandise thieves

It’s not that a Meta data center consumes too much or too little water. It is contaminating it

by

AI is quite unpopular among American citizens. According to a recent survey, the majority believe that it will have a negative impact on society. There is something in particular that has earned the rejection of a large part of the population: data centersand the truth is that they have no shortage of reasons. They pollute the airthey make raise the electricity bill and now they also contaminate the water with potentially deadly bacteria. what has happened. They count in Futurism that the authorities of Cheyenne, Wyoming, have prohibited all data centers from discharging any type of waste into the drinking water network. The reason is that Goat Systems LLC, a Meta contractor that is participating in the construction of a new data center, discharged water contaminated with Cupriavidus gilardii to the public sewage, a rare and very dangerous bacteria, especially for immunosuppressed people. The investigation. It’s not that anyone was monitoring the water quality at this particular data center, but rather that the bacteria was detected incidentally during routine testing for fecal contamination. “It’s not something we normally look at,” Frank Strong, head of the Cheyenne water agency’s engineering division, told local Wyoming News. The trail led them to the data center that Meta is building in the city, although Strong admits that they still do not know the exact origin of the pathogen within the facility. Consequences. As soon as the origin was determined, the authorities closed the facilities and stopped the discharge immediately. Additionally, as we said, they are revoking permissions so that other data centers can do the same. This means that they will not be able to do the process known as fill and flushwhereby they fill their cooling systems with water and then purge them before startup. Cupriavidus gilardii. It is an environmental bacteria that is found naturally in the soil. According to a study published in March of this year, is considered a opportunistic pathogen which rarely infects humans, but if it does it can cause death. To date, seven cases of death caused by Cupriavidus gilardii have been known and, at the moment, none have been recorded related to the spill from this data center. Meta’s response. In a public statement, a Meta spokesperson recalled that the bacteria had been found in wastewater, not public drinking water, and stated that it “immediately stopped discharging industrial wastewater and began transporting it outside the facilities.” However, the sewage system where the bacteria was located goes to a plant where wastewater is treated and then reused in public spaces such as parks, hence the authorities’ ban. What is clear is that the popularity of data centers in the US does not seem to be improving anytime soon. Image | Xataka with Magnific In Xataka | We already know how much water Amazon consumes in its data centers. We have good and bad news

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.