Model

130 meters with batteries equivalent to 550 model and

by

The transport network electrification To reduce emissions as much as possible is one of the challenges that companies and countries are facing. He electric car It is what catches us more closely, but where we also see progress is in the electric plane and in him maritime transport sector. And, in this sense, Latin America has taken a giant step with its new 100% electric ferry: the ‘China Zorrilla’. So much that it is already the largest electric ferry on the planet. Enormous. The ferry is a good starting point to start the electrification of the sea. Among its advantages is the Reduction of direct emissionsbut also lower maintenance costs (in theory) and a significant reduction in Submarine noisebeneficial for life in that environment. According to estimates, electrifying the 50 -meter length ferris to cover 100 -kilometer routes would save 50% fossil fuel in this segment, but what the ‘China Zorrilla’ has done is, as we see in MotorpasionDemential. And it is because of its dimensions. It measures 130 meters long, 32 wide and has capacity for 2,100 passengers and 225 vehicles. The ship has just been thrown into the Hobart shipyard in Australia, where it has been built, and the idea is to communicate the Argentine cities of Buenos Aires and Colonia de Sacramento, in Uruguay. It is a journey of about 55 kilometers that will cover three times a day. To batteries. The electrification problem is the same, whatever the vehicle: batteries. The larger and more capacity need, the larger and more heavy the batteries will be. In the case of ‘China Zorrilla’, these batteries have about 5,000 modules with a weight of about 280 tons. It is an absolute barbarity and, to put this in context, its creators They affirmed which is equivalent to batteries of 850 electric cars. It is the largest battery installed on a boat and its total capacity is 40 MWh. Let’s put the battery of the Tesla Model and Long Range with its 78.4 kWh. They are 0.0784, and the rapid calculations is that the ‘China Zorrilla’ has the capacity of almost 550 model and long rage. It is crazy for a 55 -kilometer journey, but in the end it is the ‘price’ of moving such a monster. And the load? Another mocked. It is the big question that may arise. If, despite the advances in speed, loading an electric car already has a considerable amount of time, what happens if we have to load a 40 MWh battery? The answer is simple: High power electroles. Specifically, giant ‘plugs’ that will admit 15 MW charges. The idea is that these loaders are in both ports, being able to recharge during the eviction and shipping of passengers with 24 MWh capacity. In theorythe 20% to 80% recharge will take about forty minutes and it is estimated that the night load pulls renewable energy, especially in the Uruguayan port. Without rival. Due to the type of controlled journeys that make this type of ships, their electrification is a very interesting idea that is already being explored in other parts of the world. In Spain, we have the ‘Barbarian Cap‘, a ferry that joins Ibiza with Formentera and has a battery of just 1 MWh. Is owned by Balearia, a company that is already shaping A much more ambitious project: Unite rate and tangier. There will be two 100% electric ferries with a total capacity of up to 39 MWh that is also provided that are loaded in 40 minutes in the ports and that will have something very interesting to reduce times: arms in the ports that will replace batteries. It will also have diesel support engines in case things are twisted and capacity for 804 passengers and 225 cars. In this sense, the ‘China Zorilla’ seems to continue reigning in the limited segment of electric ferries. And, in case you are wondering, the name of the ship pays tribute to the Uruguayan actress Concepción Matilde Zorrilla. Images | Buquebus In Xataka | The 100% electric plane of beta technologies aspires to be very versatile: it will also have a passenger variant

The DGT has been pressing a five years against Waze or Google Maps. Your solution goes through copying the French model

by

“You don’t know who you warn.” With these words, the DGT warned of the danger of giving the alarm of a traffic control through our mobile phone. It is no accident that The article It will be published in November 2024. The article clarifies the entire phenomenon seen with the one known as “Galician Method”. Last year, a person was denounced for alert 15,000 drivers through messages in a WhatsApp group of Civil Guard controls on the road. The problem is that There is no rule in the circulation code that prevents it. In the last decade, drivers have been perfecting the notices. The DGT points out that in 2012 some applications were already alerted to radars or controls but that the first boom came in 2014 with SocialDrive and Waze. Users had, in real time, information about fixed radars (That the public DGT itself) but also of mobile phones and controls. The second boom came with the Coronavirus pandemic. With the increase in road controls, users multiplied and since then many have climbed into a car from which they no longer get off. Who aspires to skip a Civil Guard control has the tools for it in Google Maps, Waze and even WhatsApp. And that is a real problem if we talk about alcohol and drug controls but also if, for a reason for general security, a device has been mounted to stop the suspect of a crime. Trying to put containment barriers, the Citizen Security Law to try to stop those WhatsApp groups with thousands of users. It is something that the DGT itself recognizes that states that in article 36.23 it is expressly prohibited “the unauthorized use of personal or professional data or professionals of authorities or members of the security forces and bodies that may endanger personal security.” However, they recognize that it is not enough and has been pressed to carry out a modification in the Traffic Law. At the moment, all that there is a proposal for the proposition of the Law born in the Commission on Road Safety of the Congress of Deputies at the proposal of the PSOE. It requested that the text reflect the obligation to “sanction those who provide information that hinders or prevents the control work of breaches of circulation standards”. In France it already applies This new wording would serve to end current doubts. The closest thing to receive a sanction for notifying the Civil Guard control is found in article 100 of the Civil Guard GENERAL CIRCULATION REGULATIONwhere the following is exposed: The use of long -range or road light is prohibited as long as the vehicle is stopped or parked, as well as alternative employment, in the form of flashes of long -range or road light and light or crossing light, with purposes other than those provided in these regulations The problem, obviously, is that this analog acting has been disused. This has led to the DGT has press and emphasize in the last five years that their intention is to chase those who use these applications to notify the controls. In 2020, Pere Navarro, director of the DGT, already put the focus on applications during mobility week in A Coruña where he pointed out his will to “Prevent, hinder or limit” Applications like Waze. The referent can have it in France. There, the famous traffic application cannot show mobile radars to users. It is an anomaly in Europe, as they collect in the Gauling country media. They point out that there the decree No. 2012-3 of January 3, 2012, which modified its circulation code more than a decade prohibits “Possession, transport and use of radar detectors”. However, this application extends to mobile applications, something that does not happen in Spain. Here, The law was modified in 2022 To punish the single presence of a radar detector. Until then the use of radar inhibitors and detectors was punished. For three years, the single presence of these devices is also punished with Fines of 200 euros and three points of the driving card for detectors and six points and up to 6,000 euros in the case of inhibitors. What the DGT wants to get is that the notices in mobile applications are also punished and the first step would be the modification in the traffic law that would allow copying the French model. What applications like Waze do is alert “danger zones”. To skip the restriction, the application allows you to add this notice on road sections, never in a specific position. With this notice, the driver already knows that a mobile radar or traffic control will be found later but is not specified in any case to stick or where it is installed. Photo | Waze on Instagram and Pricob ioan In Xataka | The DGT denounces the breakage of seven radars: there are up to half a million euros at stake and three -year jail sentences

With O3 and O4-mini, chatgpt reason better than ever. But more important than the model is that we can ask you certain questions

by

I read Alberto Romero a few days ago in his great The Algorithmic Bridge counting Three stories who shared something disturbing: they were Three people saved by doing something as simple as asking chatgpt About his symptoms. They did not use Prompts sophisticated or advanced techniques. They only described pain, discomfort, a concern. And they lived to tell. The interesting thing in the conductive thread of these three stories is not that the AI ​​was right (although he did). Is that Natalia came up with asking Chatgpt for that tension in his jaw. It is that Cooper thought about sharing his dog’s analytics. And that Flavio, with an increasing pain, will value consulting a Chatbot In addition to a doctor. The protagonist of these three stories It is not chatgpt, but that previous momentthat moment that someone thinks “What if …?” We live obsessed with what the machines are taking us away. Employment, singularity, creativity. But along the way we are losing something almost equally valuable by our own foot: The ability to imagine unplanned uses, unlikely connectionsways of use that break the instruction manual. The gap is not only between those who dominate technology and those who are not between those who can reimagine it and those who use it following established patterns. Is Like when we buy a new oven. There are those who feel patiently to read the instruction manual memorizing each official function, and those who immediately begin to tighten buttons, turn dials and create combinations that perhaps or the manufacturer conceived. Both will end up cooking, but it is the seconds who will discover that putting the grill A medium power while turning the fan, an impossible finish is achieved to achieve the official recipes. True innovation has always emerged from those who consciously ignore theoretical limitations. The good thing is that, in the same way as Talent is very democratic, This capacity does not understand credentials. A programmer senior You can get stuck by writing code with Chatgpt while a teenager discovers that asking him to simulate difficult conversations before having them. And it turns out that it works as an emotional mirror that never appeared in any manual. The stories that Alberto collects have something more in common: they occur in the medical field, where knowledge has historically been power. Where the diagnosis flows from top to bottom. Where the informative hierarchy is so rigid that it has its own uniform. However, there are Nature, Cooper and Flavio, investing the order, leading to the expert a diagnosis that they could not have formulated but that the AI ​​could detect. It is not a matter of blind faith in technology (Everyone confirmed with professionals). It is something much more subversive: a silent rearrangement of who has access to what information and at what time. While we keep wondering if the machines are going to steal our imagination, the reality is much simpler: The decisive is not what AI takes us away, but what we think about doing with it. There is the true competitive advantage of the 21st century: in that ability to see possibilities where others only see tools with predefined instructions. Natallia, Cooper and Flavio did not save their lives for having access to superior technology, but for daring to think differently. In Xataka | O4-mini is much more than another model of AI. It is the Tesla Model 3 of OpenAI Outstanding image | Xataka

Tesla had opted a good part of his year to a Tesla Model and below 40,000 euros. Right now is an illusion

by

The year 2025 has not started for Elon Musk’s company as expected. In fact, it has begun in all the ways that could not be expected if you are in charge of a car company. At the fall in the price of shares one has to add more or less continuous attacks to the company itself with boycots of all kinds (some violent and very dangerous) and a Decrease in really alarming sales. Although it is about to see to what extent sales have been weighed down by the hangover to modernize the production lines to get the new Tesla Model and Juniper (which also came out in its most expensive version) and it is foreseeable that they grow over time, the truth is that the thermometer In countries where more electric cars are sold He has jumped the alarms. In France and Germanythe sales of his Tesla Model and have been very poor. And that is a problem for the company since in both cases we talk about countries where activists have been more belligerents against Tesla and where Elon Musk’s political approach with the extreme right has been more criticized. Part of the strategy to refloat sales was to put on the market a small and cheaper tesla. It has all the meaning of the world since SUVs are still the best selling models and the company needs an alternative that moves below 40,000 euros, with competition squeezing strongly. This tesla model and will be called, according to The Wall Street JournalTesla Model Q. The newspaper assured in December 2024 that the company’s plans went to launch this model in the coming months. Of course, it would not be the $ 25,000 car that Elon Musk has promised, this option would be more than $ 37,000. However, Reuters It ensures that these plans have been delayed and that it will not be until 2026 when we see this model and on a small scale. Better for 2026 The news agency pointed out a few weeks ago that this Tesla Model and small would occur in the United States but that there were also plans to produce it in Europe and China, so it is expected that the assembly lines could be both products at the same time and, therefore, share a good part of the components. Then, in March, Reuters He pointed out that internal sources of Tesla estimated the cost savings of production of this new electric car at 20%. It is not known if that cost reduction would be associated, for example, to a smaller battery. What the agency’s journalists assure is that the new Tesla Model Q or Tesla Model and in more affordable format would be little affected by tariffs. They point out that Tesla has been promoting the internal supply of their pieces in the United States for two years and, therefore, it would have less components affected by the new commercial barriers applied by Donald Trump that tax with 25% the import of cars but also of the pieces arrived from outside although the car is manufactured on US soil. However, and although they expected to produce 250,000 vehicles in their plants in the United States, Tesla has delayed the plans to 2026 but it is not clear what are the reasons that have led to making this decision. Keep in mind that at the end of last year the information of The Wall Street Journal He already referred to an internal document of the company shared with Deutsch Bank, so The vehicle should be very advanced In its development. It is not the first time that the company delays plans related to its lowest cost cars. For years he has been talking (and promising) that we would see a tesla of $ 25,000 but Elon Musk rejected this route during the presentation of the Tesla Cybercab And, even, this possibility of “Absurd idea”. The company’s owner has prioritized autonomous driving, ensuring that this source of income will be key to Tesla in the coming years. However, the competition is tightening and already has cars like the KIA EV3 or the Renault Mégane E-Tech (among others), vehicles that would fight with that future Tesla Model q or Tesla Model and in miniature for price and size and that are already sold for less than 40,000 euros. Photo | Xataka In Xataka | “Cheaper than a BMW 330i”: the owner of a Tesla Model 3 reveals how much has spent five years with her electric car

O4-mini is much more than another model of AI. It is the Tesla Model 3 of OpenAI

by

The last launch of OpenAI, the models O3 and O4-minihe says much more about his strategy than he seems to the naked eye. The greatness of this proposal-especially that of O4-mini-is not in Benchmarks nor in technical capacities, which also; but in the tactic that raises and that Remember a lot to which he made Tesla In a giant: First came Model sa luxury product that demonstrated what was possible, but accessible to few. 80,000 dollars. Then came Model 3surprisingly powerful but at a price that greatly expanded the market. 35,000 dollars, less than half than Model S but maintaining 80% of its key benefits. With O3 and O4-mini, OpenAi has executed a very similar movement for the world of AI. O4-mini costs just 1.1 dollars per million Tokens input and $ 4.4 per million departure. And offers a performance that rivals his older brother: 68.1% in Swe-Bench compared to 69.1% of O3which costs almost ten times more. This price-reference equation is no accident, but the aforementioned strategy that we have seen before. A “economic” product that democratizes premium capabilities and expands its potential market. O4-mini exceeds Claude 3.7 Sonnet (62.3% in Swe-Bench) costing much less. It is the best model in Aime 2024 and 2025 (elite mathematical contests) surpassing even major systems. Tesla analogy is reinforced when we see that O4-mini allows a user of Chatgpt plus ($ 20 per month) access capacities that previously required Pro subscriptions (200 per month) or Enterprise (thousands). And the master coup comes from integration. O4-mini is the first “mini” model capable of using all chatgpt tools: web navigation, run python code, analyze images and generate them. With a single affordable model, Openai is absorbing use cases for which several specialized suppliers had to be paid. Concrete examples: A startup that makes slate diagrams analysis for $ 50 per month, now competes against a system that does the same for $ 20 per month, and also solves mathematics, program and seeks information. Another that sells assisted programming tools for $ 30 per month faces an O4-mini that writes better code and also visualizes data. This explains Why Openai released O4-mini when Initially he had announced that would integrate this technology directly into GPT-5. The competitive pressure of Claude 3.7 Sonnet and Grok 3 He forced them to accelerate and have shown that they are willing to cannibalize their own premium products to maintain their dominant position. This reminds what Apple did with the iPod: it was not afraid to integrate it into the iPhone knowing that in a five years It would evaporate what generated 40% of its income. Advance or die. The key difference between O4-mini and alternatives such as Claude is very clear: while others offer larger models that consume more resources, OpenAi has achieved a compact model with sufficient performance for almost any practical task. In the same way as the Model 3 It did not need autonomy for 600 kilometers when 350 were enough for the majority, O4-mini does not need the last 5% precision when it already solves 95% of everyday problems. It is not just a technological strategy, but of scale economy: With each new user, the marginal cost decreases while the competitive advantage increases. And as Tesla, when the competition tries to react by copying its approach, Openai is already announcing its next movement. Of course, as is happening to Tesla, The Chinese threat In the medium term it is unpredictable. In Xataka | Openai’s hypothetical social network does not want to connect people. Want your data to train your AI Outstanding image | OpenAI, Xataka with Mockuuuups Studio

If you are looking for a good iPhone, a lot of eye at Mediamarkt’s last offer in this model compatible with Apple Intelligence

by

Mediamarkt has launched a new campaign, this time called “April, save“, in which we can find very good offers in mobiles. One of the best is the one that has the iPhone 16since by 839 euros We are facing a new minimum historical price of the store in its 128 GB configuration. Not bad considering that recently arrived Apple Intelligence To this mobile. * Some price may have changed from the last review A minimum historical price in Mediamarkt For this price, the iPhone 16 It is, together with him iPhone 16Eone of the best ways to launch ourselves to the current generation of Apple mobiles. Among its specifications we find some interesting sections, such as your screen Super Retina XDR OLED of 6.1 inchesan ideal size for those looking for a compact mobile. Internally we have the Apple A18 which offers good performance and great power in practically all times. To this we must add that it comes with 128 GB of storage and that it is Compatible with Apple Intelligence. In addition, it has a load for MagsafeDynamic Isand and IP68 certification. As for its photographic section, by the front it comes with a 12 MP True Dept Front camera, while we find ourselves so much with a main 48 MP sensor as with an ultra wide -angle sensor of 12 MP. Other iPhone that we can find very cheap in the Mediamarkt outlet Apple iPhone SE (3rd Gen.) – Restored article – Article Temporary Exhibition, can contain slight brands * Some price may have changed from the last review Apple iPhone 11 – TEMPORARY EXPOSURE ARTICLE, may contain slight brands * Some price may have changed from the last review iPhone 15 Plus – Open, unused, exhibition article * Some price may have changed from the last review Apple iPhone 14 – Exhibition unit, in perfect condition. 88% battery * Some price may have changed from the last review You may also be interested in these accessories for the iPhone 16 Jetech Mate found for iPhone 16 6.1 inches, antigolpes Protection against military grade falls, translucent shear rear housing, antihuellas (blue storm) * Some price may have changed from the last review Apple Airpods 4 Active noise cancellation, wireless headphones, Bluetooth, adaptive audio, ambient sound mode, custom space audio, USB-C load case and wireless load * Some price may have changed from the last review Some of the links of this article are affiliated and can report a benefit to Xataka. In case of non -availability, offers may vary. Image | Pedro Aznar In Applesfera, Apple In Xataka | Best iPhone. Which to buy in 2025 and recommended models based on budget, tastes and quality price In Xataka | The best mobiles (2025), we have tried them and here are their analysis

The new Meta Model took a very good score at the benchmarks. Maybe too good

by

We had been waiting for the new family calling 4 artificial intelligence models for a long time. Last weekend the company finally revealed those models and Everything seemed promising. The problem is that the way of announcing them is generating some controversy and an uncomfortable conversation: that perhaps they have cheated in the benchmarks. Call 4 seems great. As soon as they appear on the scene, the new models call 4 goal surprised by their excellent performance in Benchmarks. They were second in the ranking LMARENAonly below Gemini 2.5 pro experimental. However, suspicions soon appeared, because the flame 4 version that is available to all audiences was not the same as it was shown in that ranking. Trucada version? As indicated in the advertisement As a finish line, that flame 4 version was an “experimental” that obtained a 1,417 points in LMarenawhile Gemini 2.5 Pro experimental had obtained 1,439 points. Some experts pointed out that this experimental flame version 4 was a version that cheated and had been specifically trained with data sets used in Benchmarks to be able to score well in them. We have not cheated.Ahmad al-Dahle is the head of the generative division in the finish line, and therefore is in charge of the flame launch 4. This manager has denied sharply The rumors that point to what goal would have cheated to get better scores in the benchmarks. These rumors “are false and we would never do that,” he said. But it was “optimized”. As indicated In TechCrunchin that official announcement Meta did pointed to the experimental flame 4 model that had scored very well was “optimized for conversation.” In Lmarena They indicated What a goal should have explained better what type of model had sent to include in the ranking. The same calls 4 is not so good. Some experts who They analyzed flame performance 4 with synthetic or conventional tests They already warned that performance It didn’t seem so good As they claim in goal. The publicly available model showed a behavior that He did not adjust to the quality that pointed its score in LMarena. Not quite consistent. Al-Dahle himself confirmed that some users were seeing “different quality” results of Maverick and Scout, the two flame versions 4 available, depending on the supplier. “We hope that some days are late when public implementations are adjusted,” and added that they would continue working to correct possible errors. A rare release. What a goal this model will launch a Saturday is strange, but when asked about it Mark Zuckerberg He replied that “is when it was ready.” That also the model used in LMarena is not the same as people can use is also worrying, and it may begin to distrust us from benchmarks and companies that use them to promote their products. It is not the first time that This happens Not much less, and it will not be the last one. In Xataka | Openai is burning money as if there were no tomorrow. The question is how much can endure like this

The traditional couple model is in crisis. And it has translated into more people eating alone in restaurants

by

Eating is a necessity, a pleasure … and also (in your own way) A social act. Around the tables we celebrate, remember and honor. We remain for dinner in our first events, to celebrate birthdays and promotions, the arrival of the New Year or say goodbye to that friend who moves to another city. Eating has always been synonymous with sharing, especially if we do it in a bar, where it also becomes a public act, exposed to unknown looks. Or at least so it was until now. Table for one, please. It is very simple. He arrives with any restaurant at rush hour and observing the room: There are more and more people eating alone. And not by obligation or because he does not have anyone to share dinner. On the contrary. We do it by choice, to enjoy loneliness and backs to the stigma that until not so long persecuted those who sat alone at tables that were designed precisely to gather large groups. Impressions … and data. As with most trends, there are studies and percentages that help us better understand their scope. Although finding a site to eat alone It is not always simplein the OpenTable US calculates that the reserves for a single diner 64% have shot Since 2019. Moreover, between 2022 and 2023, Resyanother New York reservation platform, calculates that requests of this type increased by 21%. They are forceful percentages and it is not unreasonable to think that they are short. After all, both opening and resy register only those who reserve in advance, not those who appear in the premises without calling before. Is there more data? Yes. And although there may be variations from one study to another, most point in the same direction. In 2024, opening calculated that the alone dinners in the United Kingdom had experienced a year -on -year increase of 14%percentage that in the case of Manchester stretched until reaching 23%. Other studies They also talk that the ‘Only Dining’ He has gained weight in Germany and Japan or that the number of Americans who recognize having recently has grown by 53% In just two decades, between 2003 and 2023. And Spain? There are some indicators that suggest that Spain does not remain oblivious to that trend. In 2022 Mapfre published A FOOD REPORT in which he revealed that, although most of the Spaniards continue to enjoy the family meals, between 2017 and 2022 the number of people who sit at the table without company, especially during working days, increased 5%. In 2020, in full pandemic, Thefork also perceived that the reserves for a single person had grew by 4% In our country. Of course, they were still a minority. In an attempt to go further in 2024 Reasonwhy did A poll Among several restaurant chains with presence in Spain that showed that, at least in part of them, it is increasingly common to see customers without company. Well because they are more or because those who already ate alone before do it now more often. Alsea, behind Vips, Ginos or Foster´s Hollywood, requires that 7% of Foster´sy customers 18% of those of VIPs eat without anyone next. And what is the reason? The right thing would be to talk about reasons, in plural. When analyzing the trend, so much The sector As analysts usually coincide in certain factors that have clearly influenced customers. Above all there are two: pandemic, with its restrictions and fear of infection; and a cultural change that leads us to look with other eyes to those who sit in a restaurant without a company. “That a person was lonely has always been might, but now that taboo to exhibit loneliness has broken,” Notice in The country José A. González, anthropologist. “It was growing before pandemic, but now it has been strengthened because we have become accustomed to being alone,” agrees Chef Lola Marín. “In addition, it was unthinkable that a woman was alone to eat or take a wine to a bar. Luckily it is now more common.” Even in networks They can be found Videos of people who presume that: to enjoy their dinner alone. Generational issue? There are those who believe that there is another factor that has influenced so much or more than the COVID: the change of mentality that accompanies the Z generation and the Millennials. Whether or not the truth is that the trend is accompanied by a transformation of Spanish society itself, in which It is increasingly common meet people living alone. If the INE forecasts do not fail in 2039 there will be some 7.7 million of unipersonal homes, 33.5% of the total. Question of loneliness (s). That we normalize loneliness does not mean that all loneliness is the same or all its positive effects. In fact it also carries certain challenges, as I collected recently An article of The New York Times in which a curious relationship is pointed out: in the US the increase of the people who eat alone has coincided with a worsening of the country’s general happiness index. The reason: not all loneliness is deliberate. The professor of the University of Oxford Jan-Emmanuel de Neve even It goes further and reflect on the implications that excessive isolation can have. “That we are increasingly socially isolated also assumes that we do not test our ideas about the world with other people,” he says. The result, in his opinion, is an amplification of echo chambers and polarization. Image | Ismail Hamzah (Unspash) In Xataka | More and more Spanish bars refuse to pay at the table. Its objective is very simple: greater rotation

will launch a model that was not going to see the light

by

If you use Chatgpt frequently, most likely you are interacting with GPT-4O, the model that is activated by default. However, it is not the only one. When displaying the model selector, several alternatives appear: from specific versions of GPT-4O Designed for programming up to GPT-4.5, options centered on reasoning and other models in the test phase that are not directly available in the chatbot. The problem is that this growing variety is generating some confusion: The nomenclature has become little intuitive And many users are not clear which model is the most appropriate for each task. Sam Altman, CEO of the artificial intelligence startup (AI), recognized the problem in mid -February and announced a road map focused on simplifying its product offer. Now we know that this plan has encountered obstacles. The original promise. Although Openai presented the O3 family in December last year, only the O3 Mini variant, focused on offering better performance, ended up reaching the public. In its new road map, the company said that the full version of O3, capable of obtaining scores similar to those of human programmers in certain tests, would not be launched as an independent model. According to the updated plan month and a half ago, GPT-5 would integrate part of the technology developed by the company, including the O3 reasoning capabilities. In theory, that would help order the models ecosystem. As part of that strategy, GPT-4.5 would mark a turning point: the latest model without “chain of thought”. That is, from there, all new OpenAI models would have integrated reasoning. There come the changes. Part of what we have just seen in the previous lines will no longer be fulfilled. And it is not a filtration or a rumor: It has been Sam Altman himself Who has confirmed it in a message posted in X. “Change of plans: after all, we will probably launch O3 and O4-mini in a couple of weeks, and then we will do GPT-5 in a few months,” he said. In summary: O3, which was discarded as an independent model, you will see the light. Click to see the original message in x The executive has also dropped some interesting details. On the one hand, as can be seen from the previous message, both O3 and O4-mini will be available soon. In addition, he confirmed that GPT-5 will not take long to arrive, presumably this year. Of course, there are no concrete dates, and it remains to be seen if in that margin of time we will attend another unexpected turn, such as the one we are witnessing now. What is behind the change of plans? The big question is what has happened in OpenAi to produce this new unexpected turn. It is not a radical change, but it arrives at a delicate time: the American company is building a solid reputation around its ads and its real ability to fulfill them. Such a movement can weaken that trust, as valued by both users and investors. In relation to change, Altman explained that “there are many reasons for this, but the most exciting is that we will be able to make GPT-5 much better than we originally thought.” Far from staying in an optimistic statement, he also admitted that integrating everything without problems was “more difficult than we thought” and that they wanted to make sure everything tied, especially because they expect “an unprecedented demand.” More challenges in Openai. It is no secret that Openai is subject to increasing pressure. The startup, which initially grew with the support of Microsoft, competes to maintain its leadership position in the development of AI and to ensure the necessary financing to sustain its ambitious rhythm. To all that is added a challenge not less: guarantee access to the computing capacity that your projects require. Images | OpenAI | Screen capture In Xataka | In Anthropic they have become obsessed with some trivial but key appearance for AI: that Claude will pass ‘Pokémon’

That a model of AI is stuck playing something as basic as ‘Pokémon’ seems worrying. It is not at all

by

It was 2013 and almost no one had heard of Deepminda small artificial intelligence startup. His researchers came up to make their AI system learn to play (already win) video games, and They trained her with some titles of the old Atari console. Among them was ‘Breakout’ (in Spain it appeared as ‘Arkanoid’), and A video of the time It shows how after 10 minutes playing the machine did not know just anything. After two hours of play, yes, I already played as an expert. But at four o’clock something amazing spent: The machine discovered a “trick” To maximize the effort: it made the ball end up creating “a tunnel” and then cast the ball through that tunnel so that it would not stop bouncing and ending almost the entire level effortlessly. Since then using video games to train AI models or to check if they are able to adapt to them and complete them is common in the industry. It is precisely what Anthropic tried when a few weeks ago Claude 3.7 launched. This hybrid model of AI has proven to be a notable advance in areas such as programming and reasoning, but in Anthropic they wanted to test it with a singular test: To play the ‘Pokémon’ video game. The AI ​​is stuck In this experiment those responsible for Anthropic wanted to evaluate whether the AI ​​systems “can face challenges with increasingly complex competences, not only through training, but of generalized reasoning.” Claude’s previous versions had a bad time even trying to start playing from the video game’s beginning screen, but Claude 3.7 Sonnet’s “expanded thinking” allows the new model «Plan in advanceremember their objectives and adapt when the initial strategies fail »in a way that their predecessors did not do. For those responsible for Anthropic these improvements will end up helping to solve real world problems. It is something we are also seeing With the benchmark arc -agi 2which is precisely aimed at measuring the ability of the Ias to do things that are easy for us (controlling a video game, solving a visual puzzle) but these models are especially difficult. Source: Anthropic. The advance of Anthropic here is remarkable, but is far from being able to be considered a success. In fact and how they comment In Ars Technicathousands of spectators have proven On the Twitch Channel created by Anthropic how Claude stayed totally stuck in Mount Sléniteone of the video game sections. In that channel you can also see how Claude is still trying to solve the problem and advance. “Think” and “reason” and even shows what “thinking” and “reasoning”, but the model still does not overcome that video game. And despite everything, this is a great achievement of AI Taking into account that the video game is oriented to children, it seems easy to despise the achievement of Anthropic, but these advances must be valued very positively. To start, Claude 3.7 model used to play was not “pressed” to play the video game: I had to learn about the march and adapt to the game. Here also Claude “sees” the screen and what happens to react based on that analysis. And the problem is that The ‘Pokémon’ graphics are very basic and pixelatedwhich raises an even greater challenge for the Anthropic model: with better graphics it would probably behave much better, explained one of those responsible for the experiment. Even so, Claude behaves especially well in the parts of the game in which text is shown, something that allows this model to better recognize what he needs to do in that phase of the video game. But if there is a serious problem, that is also that of memorization. Claude has trouble remembering everything you have learned: It has a limited “memory” Of 200,000 tokens and when they exhaust Claude, they resort to summaries and condense the information, which can lead to eliminate small details that are important to advance in the game. Be that as it may, the achievement of Anthropic remains remarkable, and points to a future in which these models can play autonomously and do so exceptionally to all kinds of games. As Deepmind already did it with that simplistic version of the ‘Arkanoid’, but in a big way. In Xataka | The latest Google is an AI that plays video games. THE KEY: DOES IT UNDERSTANDING NATURAL LANGUAGE

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.