The future will be full of robots, says Nvidia. And just launch a tools to win that race

The event GTC 2025 Nvidia began yesterday with an inaugural talk in which Jensen Huang machine -gunned us with a real Building burst. Among them, especially those focused on an area in which the company is especially optimistic: robotics. Foundational model. In March 2024 NVIDIA He already told us about his Project Grootand now said project has crystallized in the call Isaac Gr00t N1which qualify as the first open foundational model for humanoid robots. That is to say: the model is pretended, but it can be customized for all types of scenarios, both “domestic” and industrial. “The era of generalist robotics is here”. Huang declared how this new era began with this launch, and pointed out that thanks to Isaac Gr00t N1 and new data generation systems and learning for robots, developers can create much more capable robots. Robotic labor. The CEO of Nvidia highlighted how “it is very likely that this is the largest industry of all. At the end of this decade, at least 50 million workers will miss in the world. We would be more delighted to pay $ 50,000 to each one to come to work; we will probably have to pay the robots 50,000 dollars a year (instead).” Humanoid robots in sight. In The last part of your presentation Two and a half hours, the Nvidia CEO focused precisely on robotics, and for example showed videos of humanoid robots Neo gamma of the company 1x, which are based on Isaac Gr00t N1 and that already show their Potential as domestic robots. Other companies like Boston Dynamics –Creators of Atlas-, Quesoe Robotics and Neura Robotics are also developing this type of robots and have enjoyed preliminary access to the Nvidia robotic platform. Think quickly, think slowly. This foundational model uses a double architecture inspired by human cognition. Thus, system 1 is “a fast thinking action model” similar to what humans have with reflexes and intuition. System 2, enhanced by a vision model, is “a model of slow thought” that reasons about its environment and the instructions he receives and then raised his actions. Learning to move. This model therefore raises a pillar to develop robots (humanoids or not) that can operate in the real world. Thus, GR00T N1 can generalize and adapt to common tasks such as taking objects, moving them with one or two arms or carrying them from one place to another. Basic and limited operations are at the moment, but they are an important first step for the robots of the future. Basically these robots are learning to walk. A customizable model. Nvidia offers large information so that developers can start working with this model. Thus, we have both the Technical information Like project files available in Github. And data to train robots. Robots need huge amounts of data to be trained and learn to interact in the physical world. Nvidia ACBA to offer A huge and available data set in Hugging Face for this area. They are 15 TB of data that represent “more than 320,000 trajectories for robotic training”, in addition to 1,000 “scenarios” Scene Description (Openusd) to contribute to that training. Collaboration with Disney and Deepmind. Another of the robotic projects is the alliance with these companies to develop Newtonan Open Source physics engine that allows robots to learn how to manage tasks of all kinds with greater precision. Disney will be one of the first to use these robots, and in fact Huang finished the demo with a unique co -presenter: a robot that he called Blue and that was inspired by the BDX droids of Star Wars. It is something that Disney already pointed out years agoand it is likely that this type of entertainment robots will end in their attraction parks. Image | Nvidia In Xataka | We believed that Atlas was the pinnacle of modern robotics. Boston Dynamics has just killed the famous hydraulic robot

There are two clear winners in the AI ​​race in mobiles. The rest comes bad and afternoon

There is a clear career that most manufacturers want to lead: that of artificial intelligence. If there is a company that wanted to bet with all the meat on the grill for the IA that has been Samsung, with its concept Galaxy AI. Although its software has a large basis, support in Gemini Nano It is the key to a good part of its functions. This Google model is open to the rest of the manufacturers, who have been implementing these functionalities to a greater or lesser extent. Let’s tell you Who are leading the AI ​​race on mobile phonesand how the scenario for the short term is drawn, since the evolution of each ROM is beginning to be relatively predictable. Samsung, the absolute leader Samsung It is the company that is best doing with AI, no matter how much. The functions of Galaxy AI start from the basic functions of Gemini Nano, but are complemented by Samsung’s interpretation to provide their devices with even more specific and exclusive functions. Translation and transcription of calls in real time, interpreter mode, transcription of voice to text in APP recorder, Circle to Search Improved and adapted to the S-PEN, native translation in apps such as WhatsApp, and a generative engine engine using text prompts or on our own stroke. In the same way, we have the promise that apps will interact with each other naturallystarting with basics like WhatsApp. This is a function in which none of its rivals is working, and that contributes even more value to the family. Functions that arrived in the Galaxy S24 and that have been refined in the Galaxy S25 family. Today, no telephone has the potential at the generative level, with the translation functions so polished or the summary quality. Google has the model, but not the tools Google has a double -edged sword. It is the one who created the Nano model that feeds the AI ​​of most Android phones, but has been implementing the apps so long, than There are functions that we don’t even understand how AI (although they are). For practical purposes, this is even positive, and makes him take second place. Google is a leader in elimination of objects through Google photos, it makes it great transcribing voice to text, it is capable of Reimagine What has happened in our photographs, and has a fairly complete gemini integration. The Pixel continue to grant the cleanest experience, and this is related to AI: using a Google mobile is to use native functions to which we are already used to and, even if they are based on AI, we do not perceive it as something external or novel. It is simply our pixel functioning as it should. Apple arrives halfway, but it arrives Apple Intelligence already works in Spainthrough beta for iOS developers 18.4. It is currently one of the few companies that its software has in Beta: the rest has been stably. Despite this, and although its operation has a lot of margin of improvement, there are some characteristics that Apple shine in front of some of its rivals. And that does not speak precisely well about the rest of the functions that are being implemented in Android. The key is that their rivals are based on Gemini as an assistant. An assistant who, for the moment, cannot do too many things. Apple has integrated chatgpt in Siriand this is a fusion that we had not seen to date. I have to say that, over the days, the experience is being increasingly positive, even if it remains quite green. In fact, Apple itself has admitted that He is not ready to implement the new Siri in his AIpointing out that until 2026 we will not know her. The rest of the functions are quite similar to what the Gemini model proposes: summaries, transcripts, elimination of objects and people … everything works quite behind what Samsung achieves and, for the moment, there is not a single function in which I shine above its rivals. The key with Apple is that, at some point, Siri will be able to read everything on the screen to understand the request we are looking for. What are Chinese manufacturers Oppo, Xiaomi, honor and manufacturers as living have quite different interpretations. Within China, Oppo is leading the AI ​​career. Oppo introduced its Oppo ai with reindeer 12and I have to admit that it works quite well. The elimination of objects is one of the best we have seen to date, it is able to transcribe voice to text and has useful functions, such as detecting people with closed eyes in photographs to “open them” by AI. The rest of the functions are quite lower, such as the elimination of reflexes in photos, increased artificial sharpness, and generation of avatars by AI. Xiaomi arrives quite late with Hyperaiimplementing the functions that we have already seen in the rest of its rivals, but with a fairly poor operation. So much so that it is not even good to eliminate people from the photos, something that Google photos for a few years. Honor has Yoyo as AI agent, but in Spain he does not perform too many functions. In China, this agent now works with Deepseek and understands natural language, pointing ways. Knowing the state of the career in which they meet, Honor has invested More than 10,000 million dollars For his Honor agent aia commitment to the understanding of natural language within the native apps themselves. It is still in the development phase. Image | Xataka In Xataka | What is artificial intelligence: a simple and complete explanation for everyone

To win the AI ​​race, OpenAi wants the US to forget about laws. Specifically, those of Copyright

OpenAi is immersed in one demand seriesand all of them for the same thing: the alleged violations of the copyright that he has committed when training his AI models. Now a unique idea has occurred to get rid of all those problems. No copyright for AI. In one Proposal published by OpenAIThe company suggests the US government to consider a “copyright strategy that promotes the freedom to learn” and that “preserves the ability of American models to learn from materials with copyright.” Or what is the same: that copyright laws are not applied. IA companies have done what they wanted. We have been in which the demands for copyright rape to AI companies have been frequent. Companies that develop these models have shown no shame in this regard, and The funny thing is that there is still no consequences. What OpenAi now asks is that there are definitely not and that those works can work without legal concerns. China steps on our heels. The main argument to recommend something like that is to compete with more guarantees against China. The Asian giant has demonstrated striking advances, and in fact in Openai indicate how “although America maintains its leadership in Ia today, Deepseek It shows that our leadership is not broad and is narrowing. “ Fair use. As usual, the excuse of a “fair use” of the contents with copyright appears. According to the proposal: “If the developers of the People’s Republic of China have unlimited access to US data and companies, they run out of fair use, the AI ​​race will have finished. United States loses, just as democratic AI does. Ultimately, access to more data from the widest possible range of sources will guarantee greater access to more powerful innovations that provide even more knowledge.” “AI Action Plan” in sight. In January Trump revoked The Executive Order on the Biden He had signed In October 2023. Shortly after he issued a new one And he proposed an “Action Plan” that should be ready in 180 days. Openai’s intentions are that this plan includes such concessions. But the relationship with Trump is delicate. It is true that Openai is the great fencer with softbank of the Stargate projectand that is an initiative that Trump has presumed a lot. However, the relationship of Sam Altman’s company with the current president is complex, especially since Openai is In full legal battle With Elon Musk, the main advisor Trump. White letter. In Openai they seek to have white letter to train their models with works protected by copyright. Not only that: they also want to get their tools to help modernize government agencies being approved more quickly. Experts have not been warning that a too premature adoption of these tools could have dangerous consequences, for example in terms of possible leaks and information security. OpenAi commercialized Chatgpt Gov In January precisely with the idea that government employees had access to this type of services. China’s AI models should be revealed. The company’s proposal led by Sam Altman goes further and indicates that the AI ​​models of the People’s Republic of China are prohibited. According to them, models like Depseek They are “financially supported and controlled by the State”, and pose national security risks. Companies such as Microsoft, Perplexity or Amazon stay on their servers from the Deepseek service, but the data stay on US servers, so it seems difficult for the China Government to have access to them. Image | Flikr (Techcrunch) In Xataka | 5,000 “tokens” of my blog are being used to train an AI. I have not given my permission

Openai steps on the accelerator in the AI ​​race with its new and advanced language model

The career for the development of artificial intelligence does not stop accelerating. In November 2022 we witnessed the launch of GPT-3.5the model that laid the bases of Chatgpt and marked a turning point in the The conversational. Almost two years later, we are facing the arrival of GPT-4.5a new evolution that aims to continue expanding the limits of this promising technology. We are facing a transition model between GPT-4O and GPT-5, but with a key role in OpenAi’s strategy. The company has decided that GPT-4.5 will be the last without incorporating “chain of thought”, giving way to a new generation with incorporated reasoning capabilities. This change seeks to improve the clarity of your family of products, that has become complex over time. A more advanced model, with greater emotional intelligence AI is reaching a point where we no longer settle for having models capable of solving complex problems. That improve in programming, science, engineering and mathematics is very good, but we look for models that can interact with more spontaneous way With humans. This makes us understand OpenAi with GPT-4.5. His new language model is presented as the greatest and most advanced published by the startup to date. The researchers affirm that it is not only more natural, but their knowledge base has expanded to face complex challenges in various scientific disciplines, while improving their ability to solve logical problems. GPT-4.5 wants to stand out for its emotional intelligence, and in a recent test it has proven to live up to it. A member of the OpenAI team tested his abilities by sending him a message full of emotion. With a natural tone and Without hiding your anger, He explained that a friend had canceled him at the last minute and asked for help to write a message that reflected his discomfort. Chatgpt with GPT-4.5 detects that the user is frustrated and, instead of enlivening the anger, he suggests a calmer and diplomatic message to express his discomfort without damaging friendship: “Hello, the truth is that it has annoyed me a lot that the plans have been canceled again, I really wanted to see you. Can we talk about what is happening?” To achieve this change, Openai has modified its focus on creating models. Not radically, but incorporating New supervision techniques Together with traditional methods such as the fine supervised adjustment (SFT) and learning by reinforcement based on human feedback (RLHF), which were already fundamental in the development of GPT-4. In development. Images | OpenAI In Xataka | Alexa+, first impressions: the explosion of AI feels great to the Amazon assistant. But there are also many unknowns

Alibaba had been left behind in the Tech China race. Now you will invest 52,000 million dollars to earn the AI

A little less than a year ago Joe Tsai, co -founder of Alibaba and current president of the company, gave a worrying fact: China had a two -year delay compared to the US in AI. That disadvantage seems to have faded completely in recent weeks, but the company chaired by TSAI is not a protagonist in that segment. It is precisely what you want to change. THE TROZÓN DE MA. The company, one of the most important in the world in the technological segment, – and The 29 of the global ranking By market capitalization – it has had somewhat difficult years After what happened with Jack Ma. Now he has a clear plan to recover lost time. But Alibaba gets serious. The firm advertisement on Monday that plans to invest at least 380,000 million yuan (about 52.4 billion dollars to change) in cloud computing infrastructure and artificial intelligence. He will do it in the next three years. It will have (a lot) competition. The announcement occurs just at the time the AI ​​segment in China It is especially hot. Bytedance, owner of Tiktok, hopes to have a capex of 150,000 million yuan only in 2025 (20,690 million dollars). Others such as Tencent, Baidu or Startup Deepseek They are certainly putting very interesting things in this area. The gold fever of the data centers. In recent weeks we have seen how technological companies in the US have announced astronomical figures for their budgets and their capital expenses (CAPEX) in the coming years. Except Appleall Big Tech will make colossal investments ranging from 65,000 million finish to the 100,000 that plans to invest Amazon. Most of that money will go to data centers for AI, and here Alibaba seems to not want to be left behind. With Xi Jinping’s blessing. The Chinese government has been maintaining its great technological government very at bay. However, the Recent meeting of President Xi Jinping With the top responsible for these companies, he has raised a change in Chinese policies, now apparently more open than their private companies – although closely linked to the Xi Jinping administration – grow remarkably. Up to rise. Alibaba’s announcement occurs days after their financial results, which were slightly above expectations. That has been a revulsive for their actions, which rose significantly and that seem to reflect the optimism of the investors, which is probably even greater. Image | Alibaba Group In Xataka | While all looks were heading to the US, China silently developed a very potent ecosystem AI

The new space race has created Boomerang scrap. The probability that clash against a plane has also increased

We are launching more rockets to space than ever. Between China and Spacex, they occur releases every few days in a new space race with multiple objectives on the horizon. It’s something that is lowering space transportationbut also generates a new problem: the amount of scrap that orbits our planet. And this generates another conflict: with more satellites and rockets in orbit, the risk that a fragment of some of them hit a plane is increasingly high. At the University of British Columbia they have proposed to analyze it and have determined that, to anyone’s surprise, the consequences of the clash would be devastating. The problem. Beyond the test rockets, transport of goods and people, something that is causing more and more launching are the satellites that bring the internet to any corner. There are several players in this segment, but Spacex carries the front with its program Starlink. These satellites are sent in rockets that, when they fulfill their function, lose height and return to the planet. Some partially disintegrate and others, to the re -enter the atmosphere without controlthey fall anywhere. The ocean is usually the main receptacle, it is also possible that they fall in urban areas or that, on their way, they clash against a plane. The probability. First of all, tranquility: the risk that these space debris impact an plane is still low, very low. According to The Aerospace Corporation, in 2021 (when they were thrown, but it was not the current fever), that risk was one between 100,000, or 0.001%. The system predicted, taking into account future releases, which by 2035 would rise to seven out of 10,000, or what is equal to 0.07%. It is, as we say, a low probability, but that is there. In the study From the University of British Columbia they have analyzed how all this depends on air traffic density. Taking as an example the traffic of September 1, 2023, and the United States as a area, we can see that, every year, there is a 99% probability that the resentments of rocket bodies occur in green areas, 75% in the yellow, 26% in oranges and 0.8% in red. Las Rojas are the most activity areas, such as the main airports, oranges are large cities and green and green move away from urban nuclei and, therefore, from the agglomeration of airplanes. Long March 5B. Beyond the direct clash of this space scrap against a plane, something whose probability remains exceptionally low, there is another problem: the danger of happening and leads to the decision to cut the airspace. On November 4, 2022, the body of the Long March 5B rocket, 20 tons of weight, re -entered the atmosphere, falling on the Pacific Ocean. All good, but it might not have been like this: the entrance location was the product of chance, since the body of the rocket was abandoned in the orbit and a planned design was not made for the re -entry of the remains. Consequences. The night before the reentry, different surveillance agencies, as well as the European Air Safety Agency, issued reports in which they encouraged national authorities to restrict airspace “in a corridor of at least 70 kilometers and up to 120 kilometers on each side of the estimated reentry trajectory ”of the rocket. Spanish and French authorities complied with this and They closed part of their airspace. As a result, 645 flights were delayed, with an average of 29 minutes per plane. In addition, some airplanes that were in full flight had to return to the origin airport or take a detour. It was the evidence of a lack of planning, anticipation and control over this space scrap. Interestingly, Portugal, Italy and Greece did not make the decision to close, generating other problems in their airports due to the unexpected increase in air traffic due to deviant flights. In purple, the airspace closed by the Long March 5B. In blue, his career and fall Solutions. Unfortunately, although this is a problem that will go more, controlling the reentry of space debris is not something that has an immediate solution. The researchers propose that those responsible for launching rockets also invest in controllable reentry technologies so that they do not enter the atmosphere unpredictably. These technologies include engines capable of re -effective to partially direct the rocket, but also a better mission planning so that the rocket falls into a remote area of ​​the ocean, far from populations and, evidently, air traffic. The problem is that, although the technology is there, they estimate that less than 35% of the launches perform these controlled inputs and, with 2,300 bodies in orbit with an annual increase of between 30 and 40 bodies, the risks will continue to increase. In 2001, the titanium engine coating of the third stage of a Delta 2, with a weight of approximately 70 kg, landed in Saudi Arabia, about 240 km from the capital Obviously, it is also a huge money expense, so achievements such as catch the Starship propeller and the advances of Spacex engineers so as not to have launch and throw rockets They are so important. And a global protocol is also necessary to manage these resentments and make coordinated decisions, not to delegate everything in an aeronautical industry that is not responsible. In the end, it is much more likely that there are alterations in air traffic due to this space garbage than to the clash of one of these remains with a plane, but whenever this probability is not zero, solutions must be contemplated. Images | Nature, Spacex In Xataka | The capture of the Super Heavy changes everything: Spacex has gone from being 9 years ahead of the industry to not having a rival

Deepseek has given the starting gun in the race for a cheaper AI. And China starts with advantage

Deepseek echo resonates three weeks after stirring the foundations of the industry of the artificial intelligence (AI). And possibly will continue to monopolize attention for much longer. After all, this model created by the Chinese company High-Flyer has been transformed In a symbol for China. In this Asian country there are dozens of companies involved in the development of AI, but none has so far made so much noise in such a short time As Deepseek. Also, according to Reuters Several experts in this sector have predicted that the open source nature of this model and its low access rates through its own API (application programming interface) could promote the adoption of AI and the development of applications for this technology for this technology . In Deepseek practice it has the ability to help Chinese companies cope with better Export restrictions of the avant -garde GPUs produced by US companies, such as NVIDIA or AMD. Huawei and other manufacturers want to expel Nvidia from the Chinese market Nvidia is losing competitiveness in China. It is inevitable. The H20 GPU on paper is much less capable than the most sophisticated chips that Jensen Huang’s signature currently sells. At the moment the Ministry of Industry and Information Technology of China has not officialized anything, but the administration led by Xi Jinping has restricted since last August the purchase of GPU H20 of Nvidia by Chinese companies. GPUs for Chinese are appreciated by companies such as Bytedance for their good performance in inference tasks On the other hand, Huawei has lists its own GPU for AI, the chips ascend Ai, for more than five years. During this period of time it has been refining them and increasing their abilities with the purpose of matching or even overcoming the performance of the chips A100 and H100 of Nvidia. “Training is important, but it only happens a few times. Huawei focuses mainly on inferencewhich, ultimately, will give us access to more customers “, Ensures Georgios Zacharopoulosa senior researcher of AI who works in the Huawei laboratory in Zurich (Switzerland). During the last weeks Huawei and other Chinese chip designers for AI, such as Hygon, Enflame, Tsingmicro or Moore Threads, have declared that their solutions will be compatible with the models of AI developed by Deepseek, although for the moment they have barely given more details. The interesting thing, and this is the authentic trick of Chinese companies, is that their GPUs, especially Huawei Ascend 910b chips, are appreciated by companies such as Bytedance for their good performance in inference tasks. Inference is broadly the computational process carried out by language models with the purpose of generating the responses that correspond to the requests they receive. Anyway This declaration of Lian Jyechief analyst of the omdia technology consultant, expresses very well Why Deepseek is a great opportunity For Chinese GPU manufacturers: “This development is very aligned with the abilities of Chinese designers (…) GPUs for Chinese and require a more localized and specific understanding of the industry. “ Image | Huawei More information | Reuters In Xataka | The Nvidia pulse and US administration becomes more virulent. The B20 GPUs for danger

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.