chatgpt

OpenAI wants to turn ChatGPT into a super app. Users fear the worst

by

Internal statements cited in The Financial Times reveal how OpenAI is preparing what could be the biggest change for ChatGPT since its launch in November 2022. The company It already has 1,000 million users of the free version of its models, but wants to increase the number of those who pay, and the key is the change of approach. chat is dead. The summary of the approach is in the phrase “Chat is dead”, uttered by a senior company official under anonymity. Keeping 1 billion users using the chatbot for free requires enormous computing power and therefore money. That does not seem to have a clear return at the moment, so the company no longer sees ChatGPT as the final product, but as a gateway to hook the user and gradually convince them to use the company’s paid services, such as ChatGPT Plus. A super app on the horizon. The objective, say sources close to the company, is to launch a super app that combines both programming tools and AI agents, which will make it possible to add paying subscribers to a platform that needs to eat income. Especially considering that its IPO is imminentjust sent documentation to the SEC to prepare for that move. Codex as the center of everything. The idea here is to turn Codex into that revenue engine that ChatGPT has not been. Following the launch of the desktop application in February 2026, Codex has already multiplied its weekly active user base by six, and now exceeds 5 million. While ChatGPT has a small proportion of paying users, the vast majority of Codex users pay a subscription. Third-party apps and services. The ChatGPT interface redesign is expected to begin rolling out in the coming weeks on both the web and mobile apps. In a first phase ChatGPT will “direct” users to third-party services such as Canva or Booking, they say in the FT. The idea is that over time OpenAI will get rid of prompts so that its models understand the intention of their users when they use the website or the app. Agents in power. The new approach considers that the real value of the market is not in writing poems or summarizing texts, but in using agents that help us both personally and professionally. According to those responsible cited in the newspaper, the classic distinction between a web search engine, a chatbot and an AI agent for programming will disappear so that the future ChatGPT will be crazy without us realizing it. Thibault Sottiaux, who did speak officially, confirmed that they were preparing “a personal agent who is capable of helping you in any facet of your life, whether personal or professional.” Reasonable criticism. Photo users like Reddit They have reacted with clear criticism to this news. Existing ChatGPT Plus subscribers enjoy nearly unlimited conversational access and, separately, “credits” via the Codex programming API. If everything is merged into a new super app, these users fear that this theoretical unified agent will end up consuming an account’s tokens much faster and the pay-per-use model will harm them all. If the evolution of these models has taught us anything, it is that In fact, agents have made using AI (quite a bit) more expensive for intensive users. Mass adoption is no longer enough. When the AI ​​race began, OpenAI seemed to be happy to attract the largest possible volume of users even at the cost of putting revenue at risk. They believed that they would end up capturing that part sooner or later, but Anthropic appeared on the scene. Amodei’s company has managed to attract paying users – business users – and now OpenAI sees how its initial strategy does not seem to work. In Xataka | Anthropic’s IPO is very similar to the one Netscape carried out in 1995. That is worrying

Comparing Apple’s AI to ChatGPT or Claude is a mistake. Apple is not playing that game

by

to whom They rule out Apple in the AI ​​raceeye. The company may have arrived late and it certainly may have little to show today, but its evolution over the last three years reveals three interesting things. The first, that Apple does have its own AI models. The second, that they are very far in performance from the best of OpenAI and Claude. Third, that may not matter at all. Three years of evolution. The trajectory of the technical documents shared by Apple in recent years reveals a series of more than relevant changes. In 2024 its initial proposal was limited to small models of about 3,000 million parameters (3B) specialized in solving basic tasks like generating Genmojis or text summaries. In 2025 the company launched its MLX framework to the developer community to facilitate the integration and use of local models. Now, in 2026, They propose a hybrid infrastructure based on a basic principle: Simple requests: they run in small local models on the device, you don’t even need an internet connection Complex requests: the system delegates the task to be processed in the cloud privately through Private Cloud Compute A (maybe) great idea: NAND can help. The most relevant milestone of Apple’s new approach lies in the design of its AFM 3 Core Advanced model. In today’s mobile phones we have a big bottleneck with the execution of capable (large) AI models because these devices have a very limited amount of memory (12 GB on some iPhones). To be able to fit a model with 20,000 million parameters (20B), Apple has decided to store that model in the internal SSD unit, not in memory. In the AFM 3 Core Advanced model the “experts” are in the mobile’s SSD. They are preselected and loaded into RAM to be used dynamically, optimizing model execution. Experts by prompt, not by token. It then activates a series of pruning techniques (Instruction-Following Pruning, or IFP) to activate only between 1,000 and 4,000 million parameters in a sparse manner (sparse), somewhat similar to what is done in models with Mixture-of-Experts architecture. But Apple selects these experts at the beginning of each prompt, not token by token, which allows it to avoid the slow bandwidth of the mobile’s NAND storage compared to its RAM memory. Privacy by flag. If for something Apple’s approach stood out from the beginning It was for his privacy.which is implicit when using local models. But if the request is complex, the system redirects it to the AI ​​models in Apple’s cloud, the Private Cloud Compute (PCC). Unlike other platforms and infrastructures such as those of OpenAI or Anthropic, conversations with Apple’s AI are encrypted and are totally private according to the company: this data is not shared with third parties (because not even Apple can see it) and it is not used to train its models. Five models with the help of Gemini. Although Apple is obsessed with total control of its products, this time had to give in and ally with Google so that their Gemini models could “show” Apple the way. The result is a third generation of models that are developed in collaboration with the Mountain View firm. We have five models in total: AFM 3 Core: 3B parameter dense model AFM 3 Core Advanced: sparse model of 20B parameters with activation of 1B to 4B parameters depending on the task AFM 3 Cloud: a powerful but also efficient and fast model that runs on the Apple cloud. ADM 3 Cloud (Image): for generating and editing images, the heart of both these options and the new Image Playground AFM 3 Cloud Pro– Apple’s most powerful cloud model is for autonomous agents. It has been trained with Google TPUs and runs on Nvidia GPUs within Google Cloud infrastructure Performance, an unknown. Unlike what other companies usually do when they present their models, Apple has not published metrics on known benchmarks. Instead, it shows “human preference” metrics in which it compares user satisfaction when using its models versus competing models. The comparisons are also with previous versions of these models, which does not clarify much what can be expected from them. But they are not in the race for the best model. In 2025 yes there was comparison with open weight models of that time (Qwen-3-4B locally, GPT-4o or Llama 4 Scout in the cloud) and then they seemed to be at a good level in reference to those options. Expect them to be behind the most recent models from OpenAI, Anthropic, or Google itself, and it’s unclear how they compare to the new Chinese open weights models. One thing seems clear: Apple is not very interested in having its own Mythos, at least for now. Your objective is different. Apple models from 2026 are “preferred” more than those from 2025. Logical, but also useless when it comes to understanding how good these models are compared to the competition. But integration is important. Apple’s big ace to compensate for this difference in capacity is that its models have full access to the user’s OS, apps and hardware. AFM models are integrated with iPhone camera sensors, notification history or local app permissions. This allows useful tasks to be carried out that an LLM that is “disconnected” from the hardware will hardly be able to replicate. Here the integration of the models with the hardware and software of the device is (or wants to be) fundamental. Beware of mediocrity. This approach focused on integration and privacy is especially striking and differentiating from its competitors, but there are risks. Among others, the product is limited by its functional capabilities compared to the competition. If local models do not solve and cloud models also do not behave reliably, Apple runs the risk of having an AI that is secure and private but technically mediocre in its responses. Siri has already been criticized for being especially stupid: Siri AI must precisely eradicate that perception. In Xataka | Apple has designed Siri AI so … Read more

Booktubers already confess that they read ChatGPT summaries. The question now is what is “reading” in 2025

by

The booktubers (social media content creators whose identity revolves around reading) are starting to shamelessly admit that they don’t read the books they recommend: they read what ChatGPT says and summarizes about them. The curious thing is that, unlike what more veteran readers would do, they do not confess it to their smartphone as something they believe they should be ashamed of, or apologizing to their followers for generating second-rate content. They count it as a productivity hack, a clever solution to the problem of having to produce content about books they don’t actually have time to read. 100 books in a week. The most striking case of this trend (that is still kicking) spiked in August 2025, when a TikTok user published a video in which He claimed to have read 100 books in a week.. The trick: the SoBrief app, which offers more than 73,500 audio and text summaries with the hook of “finish any book in 10 minutes.” The reaction on social networks was immediate: what is left of reading if what you are looking for It’s not exactly Lee’s experience.r? It was even commented that these booktubers had managed to make what Bradbury advocated in ‘Fahrenheit 451’ a reality (possibly the summary does not talk about it). It’s all invented. Although generative AI is now capable of summarizing the book we want in seconds, the Internet has been doing this function for years (in a more laborious way, of course). CliffsNotes, in fact, is pre-internet: has been on the market since 1958 publishing books that summarize other books, as an aid for students. SparkNotes, founded by four Harvard students in 1999democratized literary summaries on the internet and made them free. Blinkist, born in 2012, transferred that spirit to nonfiction essays. There is a whole geneological line which ranges from these meeting points for students who didn’t arrive in time to read the books (we had ‘The Lazy Corner’) to NotebookILM and ChatGPT, which devastates all of the above: ChatGPT is free and can summarize anything in minutes. The novelty coincides with the growing pressure on creators of literary content to give their opinion on everything that comes to market. The perfect storm. Second-hand identities. Beyond there being influencers more or less honest with their followers, the conversation and the underlying controversy affects the cultural identity of the books. In the column cited above, Marc Watkins talks about the importance of the bookshelf that was seen in Zoom video calls during the pandemic (which led to the trend of hiring services that sent you books with the “right” authors for the background of your meetings). We have reached the point where the idea of ​​being readers is valued more than the act of being one. There is thousand incarnations of this idea: books sorted by color on Instagram, hauls of visits to the bookstore that are never read, the videos of “books that changed my life” with recently purchased titles… being a reader is the center of these new identities, when reading itself should be. No humans have been harmed. We have a conceptual caper that rounds out all this chaos: a good part of the books that circulate in these communities were not written by any human either. According to a study from January 2026 that analyzed 844 books from the “Success” self-help subcategory on Amazon, published between August and November 2025, 77% were likely written entirely by AI models (although these assertions must also be pick them up with tweezers). The same report states that less than 4% of the authors in that sample published 12% of all titles. There are profiles that published five or more books in the period analyzed. One of the extreme cases is that of an author who published an entire series of motivational books in three days. Human participation in this entire assembly line is minimal: the content is synthetic, it is summarized by an AI, it is commented on by creators who have not read it, and the public participates in a conversation about books that no one in the chain really knows what they are about (and it doesn’t matter much either). It traffics in the shadow of books: signs that there are books somewhere, data about their existence, reactions to those data. In Xataka | There is only something as fascinating as the work of Albert Camus, his death: absurd, unforeseen and with the shadow of the KGB

How to create a pack of Chibi stickers from your photo with Gemini or ChatGPT and then use them on WhatsApp or wherever you want

by

Let’s explain to you how to create a “Chibi” style sticker packthat Japanese style in which a face is caricatured with a big, adorable head. Let’s do this using artificial intelligenceand a prompt that you will be able to use both Gemini as in ChatGPT. Once you have generated your sticker packall you have to do is cut out each one of the image and use it as you want. For example, you can use the methods we have told you to convert any image into sticker directly on WhatsApp. The positive part of the prompt that we are going to tell you is that it will make you a series of stickers defined with pre-established expressions that you will be able to edit together with the prompt. And although it was created by OpenAI for ChatGPT, you can use it by hand both in this and in any other AI that generates images, such as ChatGPT. Create a sticker pack from your photo To create a sticker pack from your photo, you first have to upload a photo in which your face can be seen well and your features. If it’s a selfie, better, because then the AI ​​can use your features to compose the pack. Once you have attached your photo to the ChatGPT and Gemini writing field, write the following prompt. You have to send both things at the same time, the photograph and the prompt. The command to use is the following: Using the uploaded photo, create a pack of adorable, illustrated chibi stickers. Clean white background, vertical format with a thick white border. Create various tender expressions: laughing, crying, sleepy, surprised, confused, eating, grumpy, cute expressions… Each expression must include a tender text, for example: Good morning! Whatee? / Huh? / I remind you! / What a dream / Wow! Approved! / Brilliant! / Hey, you! Achís! / Angry! / Huh??? / Good night :3 / Too cute / Am I cool now?! Once you send the image along with the prompt, AI will create a set of stickers from your photograph, and it will put them all into a single image. Now all you have to do is digitally cut out each one and use it to generate your sticker for your messaging app. In Xataka Basics | How to create a character in ChatGPT and Gemini to use it in all the images you make with artificial intelligence

Claude, Gemini and ChatGPT are not supposed to be used in China. It is supposed

by

In China there is what is popularly known as the “Great Firewall”a large security wall that prevents its citizens from accessing certain services. At the same time, there are foreign companies that block their services in China, this is what is happening with AI tools like ChatGPTbut the law is made, the trap is made. Gray market. They tell it in South China Morning Postthere is a whole flourishing market of services that promise to provide access to American AI models, such as Claude or Gemini, avoiding the restrictions imposed on both sides of their borders. On online sales platforms such as Taobao or Xianyu, unlimited subscriptions to Claude Code, Gemini and ChatGPT are sold with low latency and without VPN. These platforms have become a solution for Chinese developers who want to access American models to program, debug or use multimedia generation services. How they do it. Access is done through what is known as ‘shadow APIs’ which, in essence, is an intermediary. What they do is set up proxy servers outside of China and divert all user requests there, so that those external servers are the ones that actually call the official APIs of models like Claude or Gemini and then return the “masked” response as if it were a local service, without the need for a VPN or foreign payment methods. It pays for them. According to the developers cited in the South China Morning Post piece, they resort to these ‘shadow APIs’ because they simply consider them to be tools that are clearly superior to the local offering. This translates into more precise code, fewer hallucinations and less time correcting bugs than with Chinese models which, according to what they say, still invent functionalities or fail more often. In addition, these services give them almost complete access to models like Claude Opus or Gemini, with huge context windows (up to a million tokens), without having to fight with VPNs or foreign payment methods. Wiles. All that glitters is not gold and there are also advertisements that do not fulfill what they promise. Some of these services advertise full access to models like Claude, but are actually processing requests with cheaper Chinese models like Qwen or MiniMax. Additionally, there is the risk to privacy as all traffic goes through an anonymous intermediary who could do whatever they wanted with often sensitive data. Frontier Model Forum. Is a coalition formed by several AI companies dedicated to the security and regulation of border models, but in practice it is functioning more as an intellectual property defense mechanism. Recently OpenAI, Anthropic and Google announced that they were working together to curb copying of their models by coordinating the sharing of suspicious usage patterns and distillation attack detection techniques through this common forum. Image | Xataka In Xataka | The center of gravity of mobile photography has moved to China and OPPO is going for the throne with the Find X9 Ultra

If the question is whether using ChatGPT or Claude in English is more efficient and saves tokens, the answer is: yes

by

You may not have stopped to think about it, but there is a striking reality in the world of chatbots: It is more expensive to speak in Spanish with AI than to do so in English. The reason is simple: AI does not understand words, it understands tokens. And when you talk to GPT, Gemini, Claude or any other LLM, you talk to him in a language, but to understand you he first “translates” what you are telling him and converts it into tokens. And the problem is precisely that: that not all languages ​​”cost” the same in terms of tokens. There is a very simple example that we can analyze thanks to tools like ClaudeTokenizer: the word “developer”, which in English is “developer” costs few or many tokens depending on the language in which we write it and also (importantly) the version of the AI ​​model used. In the image it is clearly seen, but just in case, we summarize: For ChatGPT (GPT-4o and GPT-5) the word “developer” has three tokens (des-developer-ador), but the word “developer” only costs one. For Claude (Opus 4.7) the word “developer” costs no less than 9 tokens (2 in Opus 4.6), but “developer” costs “only” 6 (1 in Opus 4.6). What is happening here? Well then each language model uses its own “tokenizer”your “translator” from a conventional language to the token language that the language model understands. And those tokenizers favor precisely the languages ​​in which these models are created. This is how AI understands how we speak. Each word is divided into tokens, and English is understood much better. “developer” only costs one token in GPT-5, but “developer” breaks down into three. Bad news for Spanish speakers. In fact, English has become the official language of artificial intelligence, whether we want it or not. The reason is not cultural, but architectural: 95% of the training data of the frontier models (GPT-5, Gemini 3.1, Claude Opus/Sonnet 4.7…) are in that language. That makes the rest of the languages ​​”foreign languages”, and that makes it necessary to pay extra when using them, an almost invisible toll on every interaction. In practical terms, what happens when we use Spanish to talk to an AI model is simple: we use more tokens, and therefore using Spanish is simply more expensive than using English when working with a large language model. If you want to save tokens, better use English The question, of course, is how much more does it cost us to speak in Spanish than in English with ChatGPT (GPT 5.x) or with Claude Opus 4.7? It is difficult to say because each word and each phrase is a world, but the truth is that English is almost always the most “economical”. We have used one of the first sentences in this article to compare that token consumption, and by translating the sentence into different languages ​​and querying that token consumption for different models, the data is clear. It is important to highlight that these results are not conclusive, but they do make the trend clear: English is the most efficient language in terms of token consumption, but be careful, because Spanish is not that bad, and is usually the second most efficient. It is even more efficient than English in Gemini, at least according to the tool consulted. But on average, it is normal that there is a significant extra cost when using different models. A conversation with Claude Opus 4.7 is already “expensive” because it is one of the most expensive models currently, but in Spanish it is almost 30% more expensive, not to mention in Arabic, 76.3% more expensive. In fact, according to this example, the difference between Claude or GPT-4o in terms of efficiency is clear: OpenAI tokenizer is “cheaper”and although there may be differences with GPT-5.x, what seems clear is that Anthropic has preferred to “spend more” to obtain better results (or that is the objective). Gemini is even more thrifty according to these tests, and that may also have a lot to do with the quality of the answers, although that question is for another topic. We have used one of the paragraphs of this article in Spanish and translated it with Deepl into English, Arabic, Norwegian, French and Chinese to find out how many tokens the phrase “cost” in each language. English is undoubtedly the most efficient Tokenizers advance and evolve. Sometimes they do it to save us tokens, as happened with the GPT-4o tokenizer: at that time OpenAI explained how that tool used 1.1 times fewer tokens when speaking to her in Spanish but up to 2.9 times fewer in Hindi or 3.5 times fewer in Telugu. With Claude Opus 4.7, just the opposite has happened: the tokenizer has been redesigned and consumes more tokens (up to 1.35 times more, they admitted) with the aim of better processing and understanding the text. Your chatbot thinks (and programs) in English Here we must also highlight something important: although we can talk to our favorite chatbot in any language and it will answer us in that language (unless we ask otherwise), AI models “think in English”. That is to say: when you talk to them what they do is translate what you tell them and then reason in English and finally they translate their response into the language in which you were speaking to them. This consumes additional reasoning tokens, but also has some impact on latency (how long it takes to start thinking or answer the model). In complex tasks, this can clearly influence response times for the simple reason that the AI ​​model does not stop translating from “its official language” (English) to our language. This preference for English is also noticeable in the benchmarks: in the Humanity’s Last Examin which the models are asked all kinds of general knowledge questions with several options to answer, it is reasonable to think that the models They answer better in English because that exam is designed in that language. … Read more

All the founders of OpenAI have become billionaires with ChatGPT. Everyone except Sam Altman, who has no shares

by

Sam Altman is the most recognizable face of the AI ​​industry in the world. He directs OpenAI, the company that created ChatGPT and is today valued at 852,000 million of dollars. However, a leaked document during the ongoing trial between Altman and Elon Musk Due to the change in status from an NGO to a for-profit entity, it has revealed who the true investors of OpenAI are and how much their participation in the company amounts to. In the box next to his name, only three letters appear: TBD, which in English means “to be determined.” The man who leads the biggest technological revolution in recent years does not own a single share of his own company. Sam Altman works for the love of art. OpenAI was born in 2015 as non-profit organization with an ambitious mission: to develop AI safely and for the good of humanity. That Altman did not take stock then made some sense since his role was presented as the neutral guardian, the leader whose decisions were not tainted by money. A noble mission, without a doubt. But that It is not the scenario in 2026. In 2019, OpenAI’s charity structure began to become too small to compete in the AI ​​race. OpenAI created a for-profit subsidiary under the so-called “capped-profit” model, in which investors they could make profits limited. That opened the door for capital and also for executives and co-founders to secure huge stakes in OpenAI. Altman’s name, paradoxically, remained blank. Those who did get rich, and a lot. As and how I collected Forbes, Greg Brockman, co-founder and former president of OpenAI, admitted during the trial that he owns a stake worth about $30 billion for which he paid nothing. Ilya Sutskever, former scientific director, has a participation between 30,000 and 35,000 million dollars. Figures very far from the annual compensation of $76,001 that its CEO receives, according to the tax form from OpenAI. The other major beneficiary is the Sound Ventures fund, linked to actor Ashton Kutcherinvested 30 million dollars in an early phase and that bet is now worth 1.3 billion, a return of 43 times the investment. In total, current and former employees control about $165 billion in company shares. The distribution among the greats. The block of corporate investors formed by Microsoft, SoftBank, Amazon and NVIDIA, together control 46.58% of OpenAI, with a stake valued at $396.9 billion against a combined investment of $122.7 billion. Microsoft leads that group with 26.79% of the company, a position valued at $228.3 billion built from an initial investment of 13,000 million. SoftBank occupies second place with 11.66% of OpenAI, valued at 99.3 billion compared to an initial payment of 64.6 billion, which represents a profitability of 1.5 times. amazon It has 4.66% of the company, valued at 39.7 billion dollars with an investment of 15 billion and a profitability of 2.6 times. At the top of the table is the OpenAI Foundation, the original non-profit entity, with 25.80% of the company and a stake valued at $219.8 billion which, having been formed with contributions without financial compensation, technically has an infinite return. Here may lie the key to the mystery of Altman’s retribution. A calculated move. The most widespread theory is that Altman and the board of directors, which he has firmly controlled since surviving the 2023 impeachment attempt, They are simply biding their time. Once the dispute with Musk concludes, the OpenAI board is likely to retroactively determine that Altman deserves participation commensurate with his responsibility. It is likely that, as is the case with other CEOs, This remuneration is linked to milestones like taking the company public with a valuation of more than a billion dollars. Perhaps this retribution will arise from that reserved fund now controlled by the OpenAI Foundation. Meanwhile, Altman is not exactly in trouble and your personal assets exceeds 2 billion dollars thanks to investments in companies that, curiously, are very well positioned to benefit from the growth of OpenAI. Without being a shareholder in his own company, he has built a personal business ecosystem that prospers directly thanks to his success. In Xataka | “The problem is Sam Altman”: more and more voices within the AI ​​industry are beginning to question the CEO of OpenAI Image | Flikr (TechCrunch)

How to prevent AI from always being right by default and thus make Claude, Gemini and ChatGPT have fewer hallucinations

by

Let’s tell you how to prevent AI from agreeing with you by defaultmodifying your attitude to be less accommodating. In this way, by not making an effort to please you, you will get the artificial intelligence make fewer mistakes and hallucinations. To do this, let’s compose a prompt that you must add in the configuration of the artificial intelligence you use, and which serves both Claude as for ChatGPT , Gemini or any other. It will be a prompt which we will add in the AI ​​behavior configuration so that it always takes it into account. However, remember that this will not completely eliminate the hallucinationsbecause making things up is relatively normal in AI. However, since the response will not always be directed towards agreeing with you or pleasing you, you will make them reduce it a little. Of course, another thing to keep in mind is that by doing this the user experience will change. AI can get a little “edge”because you will no longer laugh thank you. Sometimes it will tell you that an idea is bad or that you are wrong, and that will not be a failure, but will show the success of the prompt. A prompt for a less complacent AI To make your AI less complacent and verify information moreyou will have to go to the settings of the one you use and go to the custom instructions section to change its behavior. There you will have to write this entire prompt, which is quite long: Always be honest, direct and rigorous. Your goal is not to please me, but to be accurate. ACCURACY AND VERIFICATION Before answering, do an internal check: is it a verified fact or an inference? If you don’t know, say so. Do not invent data, dates, names or sources. If you’re inferring or not 100% sure, use phrases like “It’s likely that…” or “My information suggests…” instead of outright statements. ANTICOMPLACENCY (Zero Bias) Don’t give me reason by default. If my premise is false or my question is misdirected, correct me before executing the task. Eliminate unnecessary polite phrases (“Sure!”, “I understand,” “Excellent question”). Get straight to the point. If my proposal has logical or technical flaws, criticize it constructively but crudely. NEUTRALITY AND DEBATE On topics with multiple points of view, present the mainstream in a balanced way, even if my question seems to seek a biased answer. AUTO CORRECTION If you spot an error in your text generation, stop and correct it immediately. PREVIOUS THOUGHT For complex queries or questions with verifiable data, briefly reason out loud before answering. For simple queries, go direct. These instructions apply to all types of queries: creative, technical, factual or personal. How to add the prompt to ChatGPT On ChatGPTyou have to enter the settings of your website or application. Once inside, go to the section Personalization. You have to put the prompt within the option of Custom instructions. You will see that the writing field is small, but you will be able to copy and paste the prompt there without problems. How to add the prompt to Gemini In Geminiyou have to click on the button Settings and helpand in the drop-down menu click on Personal context. Once inside the customization screen Gemini, press the button Add of Your instructions for Geminiand a window will open where you can paste the prompt. How to add the prompt to Claude In Claude you have to go into the settings. Once inside, click on the section Generaland you will have to write the prompt in the field Instructions for Claude. Here you can paste everything without problem so that it is always taken into account. In Xataka Basics | The best prompts to save hours of work and do your tasks with ChatGPT, Gemini, Copilot or other artificial intelligence

OpenAI expects an 80% drop in its flagship revenue. The low-cost “ChatGPT Go” is your escape forward

by

OpenAI is in trouble. More than beforeeven. In The Information indicate that internal projections for subscribers in 2026 are worrying. The users of ChatGPT Plustheir $20 a month plan, will fall from 44 million in 2025 to just 9 million this year. That represents a drop of 80%, and they want to compensate for it with their affordable subscription. It’s not clear that plan can work. ChatGPT Go as a lifesaver. What OpenAI is going to lose with ChatGPT Plus according to these internal forecasts, they want to counteract with an extraordinary increase in subscriptions to ChatGPT Gothe ad-supported plan that costs between $5 and $8. The company’s objective is for this plan to go from having the current 3 million subscribers to 112 million, an increase of 3,600% in twelve months. A terrible quarter. While The Information showed these forecasts, in The Wall Street Journal they informed OpenAI does not have the accounts in this first quarter of 2026. The company has not achieved the expected income, and has not achieved the user acquisition figure that it had projected. OpenAI CFO Sarah Frier has warned that the company may not be able to pay for its future computing contracts if revenue doesn’t start growing immediately. The accounts do not come out. OpenAI has contracted close to $600 billion in spending on future data centers, an astronomical figure that was built with all the announcements that Sam Altman and the company made in 2025. The company expects to spend $25 billion but plans to enter $30,000, a narrow margin even if everything goes well. But according to WSJ it is not doing so, and Anthropic’s popularity has eroded its position in the market. They wanted to reach 1 billion weekly active users by the end of 2025 and they didn’t achieve it, and the decision to bet on ChatGPT Go seems like a desperate response to their revenue problem… and their IPO. No one has ever grown so much. ChatGPT Go’s growth goal poses a colossal challenge. Achieving 109 million paying subscribers in twelve months is unprecedented. It took Facebook four years to get 100 million free users, and although ChatGPT achieved the same thing in two months and set a prodigious precedent, for this to be repeated for a paid subscription even extending the time frame to 12 months would be unusual. But not even for those. Analyst Ed Zitron point Because even if OpenAI achieved 112 million subscribers at $5/month on average, it would earn $560 million per month. That figure is a far cry from the $880 million per month generated by the 44 million Plus subscribers at $20/month. The difference should be covered with advertisingbut that doesn’t seem to be going as well as they expected either. Until have activated pay per click adssomething that already caused the credibility of SEO to be greatly damaged. We go public, yes or no? According to WSJ, Sarah Friar and Sam Altman disagree about whether it is advisable to go public this year given this change in the situation. Altman wants to speed it up, but Friar doesn’t think the company is ready to meet the data reporting obligations that public companies have. The problems accumulate because the financing round closed in March made OpenAI’s valuation amounted to 852,000 million dollars. If investors had known the situation of OpenAI’s first quarter, perhaps they would not have entered that round, or they would not have done so in such a notable way. The challenge of charging $20 for AI. OpenAI’s forecast is worrying. That a company that managed to popularize generative AI can only get 9 million people around the world to pay $20 a month is disturbing and says a lot about the state of the market. On the one hand, maybe people just don’t see that $20 worth it, which is bad for the entire industry. But perhaps what people don’t see is that those 20 dollars are not worth it if they spend them on ChatGPT and they do on competitors like Claude. That is even more worrying. It is clear that there is a segment of users willing to pay such a price, but today that segment is smaller than the expectation created suggested. The Pro plan will remain a rarity. OpenAI also has the Pro plan for $200 per month, and expects its subscribers there to also double in 2026. However, that will still not be almost anecdotal because less than 1% of the total number of users—the truly intensive ones—will opt for this alternative. It is evident that this will not be the core of OpenAI’s business at the moment, and the company seems to be clear about this. They prefer to leave the middle segment in the background, have a small premium segment and bet on massive volume at a low price with advertising. We’ve seen this before… with Netflix. OpenAI’s strategy reminds us of the one Netflix launched with its advertising plan. Which many criticized when it was announced has become in a overwhelming success. The company has returned us to square one: we want to pay to see adssomething surprising but it works. And OpenAI seems to want to apply the same story. In Xataka | The surprise with the new GPT 5.5 from OpenAI is not that it is good: it is that Claude looks like GPT and GPT looks like Claude

A young man has solved a mathematical problem that lasted 60 years in 80 minutes with ChatGPT. That’s the least interesting thing about the story.

by

He is 23 years old, his name is Liam Price and he has no advanced mathematical training. Even so, a few days ago he opened the Erdös problem websitepicked one at random and pasted it into ChatGPT. I didn’t know the history of the problem or who had tried it before. What he received back seemed like a good solution, and after consulting with a friend who was studying mathematics, the two realized they might be on to something special. A few hours later Terence Tao, one of the most renowned mathematicians in the world, confirmed that problem #1196 of Erdös, a conjecture about primitive sets of integers that had not been solved since 1966, had a solution. I had found her GPT-5.4 Pro in just 80 minutes. Not like that. This problem analyzed a question about the behavior of a particular mathematical sum on primitive sets, that is, sets of integers where none divides the other, when those numbers become very large. Jared Lichtman, a Stanford mathematician, had spent years on the problem and had made partial progress, but he and those who had tried before were starting from the same starting point that seemed like the right path. A novel idea. GPT-5.4 used another starting point. He stayed in the airmetic terrain and used a special function called von Mangoldt functiona classic tool of number theory known for its connections to prime numbers and Riemann zeta function. No one had thought about that approach to the problem, and as Lichtman explained when talking about the OpenAI model solution, “The LLM took a completely different route.” The achievement is real, but with nuances. Litchman praised the proposed solution by GPT-5.4, but there is one detail that has been omitted in many comments on this event: the raw output of ChatGPT was, in the words of this mathematician, “pretty poor.” This solution made it necessary for several experts to interpret it, detail it and extract from it the underlying idea that allowed the conjecture to be solved. Price didn’t know he had the solution until his friend read it, and he wasn’t sure until Tao confirmed it. The official repository of AI contributions to Erdös problemsmaintained by Tao himself on GitHub, classify the result as a solution generated in human-AI collaboration, not as a solution developed solely by AI. The distinction is important. A previous scandal. A few weeks ago Sebastien Bubeck, a researcher at OpenAI, posted on X that GPT-5 had “solved” several Erdös problems. That publication exceeded 100,000 views, but the mathematical community and also that surrounding the AI ​​industry criticized that statement. Demis Hassabis, CEO of DeepMind, called that statement “shameful.” What had actually happened is that the model I had found solutions to already solved problems on the web. Bubeck finished deleting the original tweet and tried to back down, but all this raised doubts about the validity of the application of AI to solve mathematical problems. AI and the mathematical success rate. Terence Tao and Nat Sothanaphan maintain the aforementioned record of all AI contributions to Erdös problems on GitHub. Each of the entries in that list or table is classified with a traffic light: green for complete solution, yellow for partial progress, and red for failure. In the category of completely AI-generated solutions with no known prior literature there are three green, fourteen yellow, and eight red traffic lights. However, the repository itself adds a unique comment: those who try to use AI to solve these problems and fail do not usually report it, so it is likely that AI has been applied “silently” to a large number of these problems without success, and those attempts do not appear in any table. There is a clear bias here because only successes generate headlines. Trying to measure what matters. In February 2026, eleven mathematicians created the initiative “First Proof“. In it they included ten mathematical problems that arose naturally in their research projects. For each one they included encrypted answers uploaded to a verification site, and they gave the AI ​​systems a week to try to solve those problems that had never appeared in any training data set. Preliminary results indicate that today AI models cannot overcome that barrier autonomously, and what happens is that there are still limits to what AI can really contribute in mathematics. But then, is there a revolution or not? Terence Tao offered a clear explanation as to why GPT-5.4 had succeeded where others had failed for 60 years. What had happened was what he described as a collective blockage of the mathematical community, because everyone started from the same origin because it was “the natural one”, the one marked by tradition. The AI ​​didn’t know that was the “correct” way to start, and that ignorance turned out to be an advantage. It’s not that the AI ​​was smarter, it’s that it had no biases about how to approach the problem. Now it remains to be seen if this novel way of trying to solve problems in unorthodox ways works. This will confirm whether what happened with Erdös’s problem number 1196 was an isolated case or whether a 23-year-old boy has managed to change our vision of how to tackle mathematical problems. Image | Universal Pictures In Xataka | There is a mathematically perfect way to cut a ham and cheese sandwich and it has been discussed since 1938.

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.