spoke

The elite of the open models spoke in Chinese. Mistral has just placed Europe at a level that not even the US managed to reach

by

Over the last year, the elite of open models for assisted programming, at least in benchmarks as SWE-Bench Verifiedhas spoken with a Chinese accent. Names like DeepSeek, Kimi either qwen They had settled into the top positions in testing and were setting the pace in complex software engineering tasks, while Europe was still searching for its position. The arrival of Devstral 2 alters that distribution. It does not displace those who were already at the top, but it places Mistral at the same level of demand and turns a European company into a real contender in a field that until now seemed reserved for others. League change: the technical leap that had been brewing for some time. During recent months, the open models developed in Europe and the United States had shown constant evolution, although still without the performance necessary to compete in the most demanding tests. The progress was evident, but there was a lack of a project capable of consolidating it at a higher level and demonstrating that this path could give results comparable to those of the sector. Devstral 2 in data: performance, size and licenses. The new Mistral model reaches 123B parameters in a dense architecture and offers an expanded context of 256K tokens, accompanied by a modified MIT license that facilitates its adoption in open environments. Its compact version, Devstral Small 2, reduces the model to 24B licensed parameters Apache 2.0. In the SWE-Bench Verified figures published by the companyDevstral 2 obtains 72.2%, a mark that places it in the most competitive section of the open models evaluated and that confirms its presence among the most advanced alternatives in the segment. It is reflected by a panorama concentrated in the upper part of the benchmark. Among the open models, DeepSeek V3.2 leads the group with 73.1%, followed by Kimi K2 Thinking with 71.3% and for proposals such as Qwen 3 Coder Plus and Minimax M2, which are around 69 points. At lower levels GLM 4.6, GPT-OSS-120B, CWM and DeepSWE appear, with more moderate results. In the closed commercial environment (proprietary models), the graph incorporates higher scores: Gemini 3 Pro reaches 76.2%, GPT 5.1 Codex Max rises to 77.9% and Claude Sonnet 4.5 scores 77.2%, all of them above the best brands registered for open models. What SWE-Bench Verified Really Measures and Why It Matters. SWE-Bench Verified is a test designed to evaluate whether a model can solve real programming tasks, not synthetic exercises. Each case presents a bug in an open source repository and requires a patch to pass the previously failed tests. The evaluation seeks to measure whether the system understands the structure of the project, identifies the cause of the problem and proposes a coherent solution. It is a useful and demanding metric, although limited to Python repositories and a specific set of situations that do not cover the full breadth of software work. From co-pilots to agents who act on the project. The arrival of Devstral 2 coincides with a broader change in the way of working with programming tools. It is no longer just about receiving suggestions in the editor, but about having agents capable of exploring an entire repository, interpreting its structure and proposing changes consistent with its real state. In this context, Vibe CLI appears, a tool that allows Devstral to analyze files, modify parts of the code and execute actions directly from the terminal, bringing these capabilities closer to the daily workflow of developers. Cost and deployment: what each type of user can do with Devstral. The model will be available for free for an initial period and will then cost $0.40 per million tokens for input and $2.00 per million for output, while the Small 2 version will be priced lower. Its deployment also makes a difference: Devstral 2 requires at least four H100-class GPUs, aimed at data centers, while Devstral Small 2 is intended to run on a single GPU and, according to Mistral documentation, the Devstral Small family can also run in CPU-only configurations, without a dedicated GPU. This variety allows both companies and individual developers to find a suitable entry point. The appearance of Devstral 2 introduces an unexpected element in a space where Chinese companies set the pace and where not even the United States, despite its leadership in artificial intelligence, had an open model in this high performance range in SWE-Bench Verified. Mistral does not displace those who were already at the top, but it does broaden the conversation and shows that Europe can compete in a field where it did not appear until now. It is a movement that does not alter the general hierarchy, although it does open a new margin for the evolution of assisted programming tools. Images | Xataka with Gemini 3 In Xataka | OpenAI and Google deny that they are going to put ads in ChatGPT and Gemini. The reality is that accounts do not come only with subscriptions

The POCO F8 Pro and F8 Ultra are a great change of direction for the brand. We spoke with POCO to find out what awaits us now

by

POCO launched two new devices a few days ago that mark a change of direction in its strategy: the F8 Pro and the F8Ultra. The latter represents the Chinese brand’s most ambitious commitment to enter the premium segment, just eight months after the launch of the F7 Ultra in March. As we shared a few days ago, we had the opportunity to analyze it in depth, but we were also able to have a chat with Kang Lou, head of product marketing and spokesperson for POCO Global, and Stanley Yeh, chief audio engineer, at a press conference during the launch event in Bali. Both managers gave us very interesting clues about the future of the brand, which is at its best, proof that its strategy continues to work. A change of strategy. For seven years, POCO has focused exclusively on delivering the best performance at the most competitive price possible. “When POCO was created seven years ago we always focused on one thing: performance,” Lou explained. But now things have changed. “Since the beginning of this year, with the F7 Ultra in March, we started testing the premium market. To do so, we tried to elevate the overall user experience, not just the performance. We want our users to experience good features regardless of whether we are talking about camera, battery or any other feature.” POCO F8 Pro Collaboration with Bose. One of the great novelties of the F8 Ultra is its 2.1 audio system developed in collaboration with Bose. “In the past we usually ignored the audio area. This time we collaborated with Bose because we want a mobile phone to come close to producing the real sound that humans actually hear. We worked with Bose to redesign the entire audio system, both in software and hardware,” said Yeh. Why Bose and not another brand? Asked why they specifically chose Bose out of all the audio brands available, Yeh said, “Bass is quite important for what humans actually hear. Bose has experience in those areas and they also have a lot of patents and technology for small speakers to produce big or deep bass.” In addition, he noted that “Bose has a lot of experience in psychoacoustics, about what good sound is and what kind of sound small speakers can produce.” Qaggressive tough. One of the most recurring questions in the room was how POCO can maintain such competitive prices. Lou responded by explaining that “that’s our biggest strength as a brand. We try to reduce our price as much as possible while maintaining low margins because we really do work on low margins.” LITTLE F8 Ultra Lou further added that “POCO saves a lot of costs because we only focus on online. We don’t have the retail costs, which are a large proportion when we talk about those prices.” Furthermore, regarding the Pro model, he clarified that “we can save costs because we are using the chipset that was launched a year ago. And fortunately, we are in a market where a chipset from a year ago is still very, very capable for common use and for any type of game or intensive use.” Jeans on the back. The F8 Ultra stands out for its rear finish that imitates the texture of jeans, a design decision that especially caught our attention, as did the rest of the attendees. “We wanted something really outstanding, something that when you see it for the first time you fall in love. You feel, ‘oh, there’s something different, something quite cool, something memorable,’” Lou explained. Although he clarified that “it’s not actually denim, it’s not jeans at all. It’s a completely different material made of silicone. We chose it because it’s very durable and at the same time very comfortable to the touch.” In recent years, POCO has been characterized, in part, by targeting a more youthful audience and by devices with yellow finishes. The absence of this color in the F8 Ultra has also been noted. Lou clarified that “we don’t actually stick to the iconic yellow for every phone we make. We just change the overall design language a little bit depending on the product positioning.” Although Lou assured that there are no established plans to make denim a permanent distinctive element. “If it’s something our users want in the future on other devices or accessories, we’ll try to make the effort.” The POCOs have less battery than the Redmi K90. The POCO F8 family is based on the Redmi K90which are sold in Asia. The K90 and K90 Pro Max have batteries greater than 7,000 mAh, but this battery capacity does not carry over to the POCO in Europe. Lou explained that this is “mainly for transportation reasons, but not only that. Many of our third-party partners have warehouses in local markets and those warehouses have to be certified to store batteries up to a certain capacity. In order for everyone to partner with us and get our devices, we have a limit in place.” The camera, that great pending task. Regarding possible collaborations in the future, especially in the photographic section, Lou clarified that “we have a strong cooperation with Bose and some other leading technologies in terms of camera. We are always trying to improve the camera experience compared to what we have today. If we can have more collaborations in terms of camera to strengthen the entire future experience, of course we will. But at the moment we do not have any collaboration in that sense.” The firm also does not have a large arsenal of accessories under its POCO brand, and perhaps this can be a very lucrative aspect for the brand. At the moment they do not have a roadmap in mind, so we will have to wait to find out more details in this regard. What’s coming now. POCO is at a critical moment. After seven years building a solid reputation under the “value for money” label, the brand is now trying to … Read more

Those who spoke Spanish in their troops were not soldiers, they were druglets

by

Officially, since Russia invaded Ukraine in 2022 there were three armies fighting in the front. To the most obvious in the first bars of the contest, Russia and Ukrainethe troops of North Korea, an alliance with Moscow that It was confirmed several months ago by both parties. No one could also miss that all kinds of flags were found among the armies. But this is something else: Ukraine has just discovered with surprise some legionaries that They come from Mexico. Posters in Ukraine. The war in Ukraine has become a Innovation laboratory military, and one of its most influential developments has been the massive and devastating use of FPV dronescheap, agile and ability to carry explosives to hit vehicles, fortified positions or even penetrate windows and directly attack enemy combatants. These devices, which have transformed Mobility and logistics In the European Eastern Front, they now begin to attract the attention of non -state actors with very different agendas to those of a national army. According to an investigation of intelligence onlinemembers of Mexican posters have infiltrated the International Legion of Ukraine not only to fight against Russia, but also to acquire first -hand the techniques of use and maintenance of FPV drones that could later apply in His intestine wars In Latin America. The Legion as a gateway. From the first months of Russian invasion, Ukraine created the International Legion to receive foreign combatants willing to join their forces. In this framework, intelligence services from both Mexico and Ukraine have detected indications that some Spanish -speaking volunteers, linked to drug cartels, have enlisted with the intention of absorbing technical knowledge that they could hardly obtain in another place. The reports point to that Even the Mexican CNI He sent a memorandum to the Ukrainian SBU alerting about this phenomenon, while kyiv’s counterintelligence has initiated research On several recruits associated with semi-clandestine units, such as the ethos tactical group, which operates in Donbás and Járkov. The presence of former FARC guerrillas, equally interested in these tactics is not ruled out either. Cartel’s arms evolution. Concern is logical if observed The trajectory Of the large Mexican criminal groups, which in the last decade have traveled to improvise homemade explosives to deploy arsenals almost military. About five years ago they began experimenting With quadricopters They carried small explosive loads. Then they perfected Air bombing techniques rudimentary against enemies or security forces. Learning. In recent times it has been documented The use of FPV in direct confrontations between the Jalisco Nueva Generación Cartel (CJNG) and factions of the Sinaloa Cartel in states such as Nayarit. Videos and Drones remains recovered confirm that they are already testing suicidal attacksalthough They pointed on TWZtheir tactics still seem incipient and plagued with failures. However, the entry of operators with Ukrainian training could radically accelerate that learning curve. The FPV and its criminal application. Unlike a commercial drone like the DJI, whose operation is simple, flying an armed FPV is more similar to pilot a plane: Requires vision immersive with glassesfine control and exhaustive training to achieve precision maneuvers. In Ukraine, they have established themselves Authentic Academies for Train pilots In the offensive use of these weapons, courses that posters and other non -state groups try to take advantage of clandestine. Intelligence experts on posters foresee that, if their domain of this technology will be consolidated, FPV could be used against armored vehicles improvised (calls monsters or narco-tanques), fortified ranches of rival bosses or directly against human objectives of high value, through surprise and lethal attacks. The Ukrainian experience has shown that a drone of just $ 400, well managed, can defeat and generate positions A disproportionate impact. Concern and consequences. No doubt, for kyiv, the discovery of this phenomenon has been reason for indignation. SBU authorities admit that, although the international legion was born to channel foreign solidarity in defense of the country, they now face the reality that Ukraine has become a platform for Global dissemination of asymmetric war tactics. The risk is that knowledge designed to stop a state invader is exported to criminal conflictswhere they will be used to magnify the violence of posters that already have of heavy weaponsorganized hitmen and all kinds of corruption networks. That the CJNG or Sinaloa integrate FPV in its arsenals not only raises the level of its clashes, but also introduces a micro-war dimension Technological in Latin America, with unpredictable consequences for public safety. Global Violence School. He episode illustrates a major phenomenon: contemporary conflicts have become incubators of tactics and technologies that then migrate to other scenarios. As well as Syria popularized the improvised drones In the hands of jihadist militias, Ukraine is exporting The FPV drone as a reference weapon for non -state actors. The fact that the drugs in Mexico travel thousands of kilometers To train on the battlefield more technologically innovative of the present reflects not only the interconnection of conflicts, but also the rapid democratization of access to advanced war tools. If you want also, in a world where the cost of lethal violence is low and its knowledge is globalized, the border between state wars and organized criminal violence is It blurs dangerously. Image | SGT Williams In Xataka | For months Ukraine turned a strategic city into impregnable. Until Russia added a lethal partner to his drones In Xataka | It is not that the US has warned Russia about Ukraine, is that it has just sent nuclear bombs up to 50 kilotons to Europe

If their spies spoke Basque as a secret code

by

The espionage stories They are fascinating. Video games like Spanish ‘Commandos‘They have one of their best characters in the spy and the sagas of’Impossible mission‘ either ‘007‘are sustained (At least in the first deliveries) in that work of espionage and Double agents. Something fundamental is the encryption of conversations, but … why devise a complex system so that the enemy does not discover what you speak if they exist languages ​​that do not speak so many people? In Spain there is one that is homeless: The Basque. And, precisely, one of the legends of the spies Spaniards is the use of Basque so that enemies do not discover the plans. But as usually happens, there is as much reality as a myth in this story. Felipe II’s networks. Let’s start at the beginning and one of the clearest uses of Basque in a espionage network. In the second half of the 16th century, things in Spain were uneasy. The reason was that Henry III of Navarra, king of Navarra since 1572, had religious ideas contrary to the Spanish monarch Felipe II. While Felipe was a staunch Catholic, Enrique became Protestantism. The problem is that in 1589 Enrique III of France and ‘Our’ Enrique III inherited the French throne as Henry IV. Felipe did not make a hint of grace because he did not want a Protestant to occupy the throne of a great neighboring country and, as a defender of Catholicism and Spanish hegemony, supported the Catholic League and proposed to his daughter Isabelgranddaughter of Enrique II of France, as queen of the Gallic country. He did not set the proposal and that caused a rivalry that made the Pyrenean situation complex. Lady of Urtubia. Enrique, in the end, returned to the path of Catholicism in 1593, so things have already calmed down and was recognized in full as king of France. Now, although Enrique IV’s conversion calmed the situation, the tension did not disappear at all, and our spy comes into play: Aimée de Urtubia. This woman belonging to the nobility shipment Between 1597 and 1598, at least 19 letters to the mayor of Fuenterrabía and Captain General of Guipúzcoa. In them there was relevant information about military and political events in France and Navarra during the Religion wars and allowed the Catholics of Felipe II to record the situation in France. The peculiarity is that the letters were written in Basque, which assured that they would only be “decipherd” if they fell into someone who controlled the language. And it was his native language, detail that is not less. In whatever, their letters were referred to the Secretary General for the North, which speaks well of the importance that the information of the ‘Lady of Urtubia’ had in the strategy of the administration of Felipe II. In 1598 the Paz de Vervins with which Felipe stopped intervening in French affairs. Although the agreement had not been reached, Felipe would not have interfered with more because he died that year. Second World War. There is proof of Aimée’s letters, but … the legend of the Basque among the Spanish spies comes from there? Well, not quite. Time to make a leap to Second World War. Spies are key to operations such as day d or even for war stratagems that confuse the enemy. The Enigma machine It was the great Nazi tool to encrypt messages and Basque … too? Myth. The United States, of course, considered the use of little known languages ​​to make secret communications. Instead of creating language, Americans used Native American languages ​​such as Navajo or CREE to generate safe messages. It would be rare for a German to speak Navajo, go. And, to expand networks in Europe, they also thought of the ‘Basque Code Talkers‘. This was basically reused Euskera to produce indecipherable messages for the enemy. To the Not being an Indo -European languageor the enemy had a dictionary, or would not find out anything. In fact, some publications of years ago affirm that Basque was used as a code in the battle of Guadalcanal in 1942. The problem is that it is … false. And reality. A subsequent investigation carried out by historians Pedro J. Oiarzabal and Guillermo Taberilla, demonstrated that Americans may consider the use of Basque as ‘Code Talk’, but it was not carried out. Investigating American and British military archives, they did not find that they support the use of Basque in that context. American captain Frank Carranza, grandson of Basque emigrants, would have been one of the architects of the legend, supported by the historical fact of the relationship between the secret services of the Basque Government In exile and the OSS American. Now, something that could feed the myth is that there were Basque spies involved in the conflict. José Laradogoitia Menchaca, or “Bromine”, He was a Basque pastor who worked as a double agent during the war. For whom? First for the Nazis, but then he went to the allied side, transmitting false information to the Nazis. And the investigation of Oiarzabal and Tabernilla, regardless of denying the myth, He left a pearl: Vicelehendakari of the Basque Government in exile worked as a spy for the British. In the end, Basque soldiers may speak in Basque between them and I find it funny to think that some Nazi intercepted that communication and gave mental somersaults trying to decipher the ‘code’. Image | Screen of ‘Call of Duty ww2’ In Xataka | The US landed on an empty island during World War II. In nine days it had more than 300 casualties

Michigan’s man who killed his wife as she spoke at 911 was convicted

by

A 31 -year -old man who shot his wife five times on his head while she was talking on the phone with 911 was declared guilty of premeditated murder in the first degree. After a trial and six -day deliberations that concluded, a jury from Macomb County, in Michigan, told Steven Bryce Wheeler guilty of fatally shooting Jordan Wheeler33, with a .45 caliber gun. The shooting occurred at 9 PM of October 19, 2023 within the Wheeler residence in Woodward Street, about 15 Mile Road and Beaconsfield Street, in Clinton Township. Jordan Wheeler had called 911 that night to report that Steven Wheeler had assaulted her, police said previously. While she was talking on the phone, he shot her in front of two of the couple’s three childrenSaid Macomb County prosecutor, Peter J. Lucido. The couple was in a divorce process. The Oakland Press reported that during the separation, Jordan moved with his sister. On the night of October 19 he received a call from one of his children about his father’s violent behavior. He ran to the house and Steven Wheeler received it at the seemingly furious door, shouting expletives like “Get out of my house” and “Get out of my life”. When Wheeler assaulted his wife in the kitchen, she called 911. During the call, the operator asked if there were weapons at home, and Jordan confirmed yes. While he was still on the phone, Steven took a gun and shot his wife in his head five times. In the call you can hear: “That they fuck you. I killed you, p ”#$. I shot you in the p $%& head. ” Police arrived at the scene and arrested Steven Wheeler, who now It faces the possibility of life imprisonment without probation. “This conviction does justice for an act of tragic and meaningless violence,” he said in a press release. “Domestic violence is not only a life, but it leaves deep and lasting scars in all those involved, especially in children who are stolen innocence. No child should have to withstand such a trauma, and no community should have to endure the weight of such a loss, ”he added. The attached prosecutors of Macomb County, Kumar Palepu and Kelsey Heath, took the case in the name of the Lucid office. Wheeler is expected to be sentenced by the Macomb County Circuit Judge, Julie Gatti, on Wednesday, February 26. Continue reading:– Arizona’s mother and grandmother arrest for locking a girl who lost her life in a cage.– Maryland’s man kills his girlfriend and hides his body into an impromptu tomb at home. (Tagstotranslate) Murder charges (T) Michigan

We spoke with the creators of ALIA, the 100% Spanish AI, to understand its future

by

This Monday it was announced release of ALIA language models. The initiative has been in development for years and it is now that the first fruits are beginning to be seen, still modest, but promising. To learn more details about ALIA, at Xataka we have spoken with Martha Villegas (@MartaVillegasM), head of the Language Technologies Unit of the Barcelona Supercomputing Center (BSC). This has allowed us to clarify the status of the project, its objectives and its next challenges. To compete with ChatGPT, nothing The first thing we wanted to know is how ALIA had been created, and here Marta Villegas clarified that the model is based on the Llama architecture – Meta’s Open Source model –, “but the model has been trained from scratch and with zero initial weights“. This is important because ALIA is not a Llama-based model that has undergone a refinement or “fine-tuning” process. In those cases, this expert explained, “you start from a model trained with other data and with initialized weights, and you do it to adapt that model to your needs, either because you have more data and you want it to be better or because perhaps you want to adapt it to a particular domain. But here, he told us, “the vocabulary (set of tokens) is completely different.” In other models the corpus or training data set may be mostly in English, which causes the set of admissible tokens to be calculated through English. That, Villegas indicates, would make it adapt less efficiently to other languages. That is precisely what has been sought with ALIA: reduce the relevance of English to increase the number of 35 languages ​​of the European Union and, especially, Spanish, Catalan, Basque and Galician. How ALIA has been trained The ALIA training process began with some experiments in April 2024. It is necessary because as Villegas explained, “training is not pressing the button after feeding the data and that’s it.” It had to be taken into account that MareNostrum 5the supercomputer located at and managed by the BSC, had just come into operation at full power and there was high demand to use it. MareNostrum 5 In this training process, the ALIA project has had gradual availability of the computing capacity of MareNostrum 5. Although for a short period of time they had access to 512 of the 1,120 specialized nodes of the supercomputer, 256 nodes were used for many months and since September They are using 128 nodes, “which is a lot,” Villegas highlights. During the training process, he told us, there are so-called “checkpoints”, in which it is possible to evaluate how the training process is going. These “pauses” also allow certain training data to be updated, as in fact happened in that process in which at a given moment they introduced a new corpus with high quality that allowed them to replace some data they had. This is just the beginning: it’s time to “instruct” and “align” ALIA Villegas explained to us that ALIA is a foundational model: it is not prepared to be an alternative to ChatGPT. The latter is based on GPT-4, a much more ambitious foundational model that involved much more investment. Here we must differentiate the foundational model from the “educated” and “aligned” models with which we usually interact. As this expert told us, “ALIA-40b is a foundational model that is not instructed or aligned. For a model to be a ChatGPT and understand the conversation and have a certain memory and be “politically correct,” the foundational model (which only learns to say the next token) is “instructed” by passing a bunch of texts.” Even so, the goal is to gradually consider these options. “In March, the instructed version of ALIA-40b is expected to be launched, with a first set of open instructions,” Villegas told us. These instructions are going to be subcontracted – the ones that allow these models to be instructed – and a million euros are going to be invested in that set of instructions from scratch. This data will also be published so that it is available to institutions and developers: if it has been paid with public money, explains Villegas, it is logical that this data will also be public, something that does not usually happen with other AI models from private companies. While training AI models provides guidance on how to respond and defines the context and purpose of those responses, alignment solves problems such as avoid discriminatory biasprevent misinformation or protect privacy. Precisely this lack of alignment means that using these models in this initial phase can produce responses with errors and biases that are precisely mitigated to a great extent with this alignment phase. ALIA and the competition: it is neither a rival of ChatGPT nor does it intend to be In fact, Villegas highlights, “the objective is not to compete with ChatGPT, for that we would need 5 billion dollars.” ALIA-40b “is a good model, and a chatbot can be made in the future because the intention is to instruct and align it, but that will take time.” Within the ALIA family we have the Salamandra models (2b and 7b), smaller and more modest but which already have first instructed versions. Its performance and capacity still have room for improvement, but they are good starting points for the future. It was inevitable to ask how ALIA then intends to compete with other models, both closed and developed by private companies and Open Source models. For her “There is a demand for intermediate models that each person can then adapt to their specific use case, not everyone can use ChatGPT for reasons such as privacy or use case.” Villegas also wanted to highlight how these smaller models can have exceptional performance in specific tasks, and can work at levels of security and not sharing important data. The objective is not to compete with ChatGPT, for that we would need 5,000 million dollars Not only that, he reveals: “we also took out the … Read more

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.