memory

only Chinese memory manufacturers can bring sanity back to the hardware market

by

The price of memory has become a bad joke. DDR4 and DDR5 modules are in short supply, PC manufacturers are packaging computers without enough memory and even Apple has risen 20% the price of your MacBooks and iPads blaming this component directly. The artificial intelligence (AI) is to blame, and there is no sign of the situation improving anytime soon. Samsung, SK hynix and Micron, which control around 90% of global DRAM production, have refocused much of its manufacturing lines towards high-bandwidth memory (HBM), which powers AI accelerators from Nvidia and other companies. The result is that the consumer market, the one that supplies our computers, mobile phones and consoles, has been practically abandoned to its fate. And in that void there are only two names capable of providing some oxygen: CXMT (ChangXin Memory Technologies) and YMTC (Yangtze Memory Technologies Co.). These two Chinese manufacturers have been secondary players for years and, suddenly, they have emerged as the only variable capable of stabilizing skyrocketing prices. Two manufacturers, one opportunity CXMT has made a rise that seemed unthinkable just two years ago. This company has increased its net profit more than 1,688% in a single quarter, has signed a contract with Tencent valued at about 20 billion yuan (about $2.75 billion) to supply DRAM for servers, and already holds a global share of 7.67% according to Omdiamaking it the fourth largest manufacturer on the planet and the first in China. While CXMT fights its battle in the DRAM market, YMTC does the same in the field of NAND Flash chips. This last company has gone from a share of 8% to 13% in just one year, and is preparing a new plant whose mass production will start during the second half of 2026which will place it as the third largest NAND manufacturer in the world only behind Samsung and Kioxia. In addition, YMTC has decided to dedicate 50% of the capacity of its third plant in the city, different from the previous one, to the manufacturing of DRAM, marking an unprecedented foray into the field of its compatriot CXMT. CXMT prices are already starting to look dangerously similar to those of Samsung, SK hynix or Micron However, it is in our best interest to moderate our enthusiasm. As we could see at Computex, CXMT prices They are already beginning to look dangerously similar to those of Samsung, SK hynix or Micron, and a good part of their production is still destined for the Chinese domestic market. It is important that we also moderate our expectations: the avalanche of capacity of these two manufacturers will not reach the global market until 2027, according to the industry’s own estimates. And there is another front that does not depend on the factories. CXMT is on the Pentagon’s 1260H list for its alleged ties to the People’s Liberation Army, something that has forced Apple to put pressure on the Trump Administration to be able to buy chips from him without retaliation. As can be expected, any movement in Washington can stop this escape route in its tracks just when we need it most. China is not going to solve this crisis overnight, but it is, today, the only piece on the board that is moving in the right direction. Whether it arrives on time or not will depend on both its factories and the geopolitics that surround them. In Xataka | China needs to develop a new type of chips immune to US sanctions. And your scientists have just achieved it

SK Hynix is ​​betting that the memory cycle is dead. He proves it by doing just what always killed him.

by

SK Hynix debuted on the Nasdaq with a rise of 13% last Friday, after placing $26.5 billion in the largest IPO of a foreign company in the history of the United States. And that money does not go to dividends or buybacks. He is going to build more factories. What has happened. The South Korean has sold 177.9 million ADRs at $149 eachwith a premium of 2.9% over its price in Seoul, something unusual in this type of placement (they usually come out at a discount to attract buyers). Demand has multiplied the available supply by seven. On the first day of trading, the title reached $170. With this operation, SK Hynix becomes the second South Korean company to exceed $1 trillion in capitalizationafter Samsung. In the last twelve months its stock has risen more than 630% in its home market, Seoul. Why is it important. The memory sector has been trapped in the same pattern since the nineties: Excess demand. Huge investments in capacity. Oversupply And price collapse. It has happened in 1997, in 2001 and in 2008. This time SK Hynix defends that AI has broken that pattern forever, that the demand for high-performance memory has no foreseeable ceiling. And to demonstrate it, it does the only thing that lit the fuse in the three previous crises: pouring billions into expanding factories just when margins are at historic highs. In figures: SK Hynix controls just over half of the global memory market HBMwhich NVIDIA GPUs need to train AI models. Its operating margin for the first quarter of 2026 was 72%, with a net margin of 77%. Together with Samsung, it has committed more than $500 billion to two new manufacturing complexes. The fund for extreme ultraviolet lithography equipment alone amounts to about $7.9 billion. The context. All this has a technical reason: the 16-layer HBM4. NVIDIA has asked SK Hynix, Samsung and Micron to have it ready before the end of 2026, when the previous 12-layer standard has not even entered mass production. Stacking 16 layers forces each wafer to be thinned to 30 micrometers, a third of a hair, with an almost zero margin of error: a single defect ruins the stack. SK Hynix has already shown a 48 GB module and plans to mass produce it this quarter. Samsung lags behind, with a yield per wafer of just 10% at the beginning of the year. Micron has preferred not to risk: it has already sold all of its HBM4 production from 2026 and postpones the jump to 16 layers until 2027. Whoever wins this race keeps the orders for the next NVIDIA GPU platform, where SK Hynix would already supply around 70% of the memory. Between the lines. The market says it believes in the end of the cycle, but it does not act as such. What you say vs what you do: In June, a simple comment from SK Hynix about slowing down part of its AI memory business led to the fifth-worst day in Kospi history. Micron, its reference in the United States, continues to grow a lot, but no longer at the rate of a few quarters ago. Yes, but. SK Hynix, Samsung and Micron together control more than 90% of the market and face a lawsuit for alleged price pact in the jump from standard DRAM to HBM. And all three are expanding capacity at the same time, with the Micron factory in New York without contributing real volume until 2028. When all this production arrives together, the sector will return to the usual question: what happens when supply catches up to demand. SK Hynix is ​​betting that, with AI, the answer is no longer the same. In Xataka | 0.3 nm chips have a definitive date: the IMEC laboratory has found a way to extend the life of silicon Featured image | Xataka

If the price of memory in 2026 is high, in 2027 it will be apocalyptic. And one of the three that manufactures that memory says it

by

When we talk about what cell phones are more expensivethe consoles continue to rise in price or that to buy RAM memory you almost have to take out a mortgage, we have to know that three companies (Samsung, SK Hynix and Micron) control around 90% of world production and that they have abandoned the consumer sector to focus on memory for AI platforms. This is very bad news for consumers, but excellent for those companies that are seeing record profits. And if the question is when the storm will end, the answer is that we are in the eye of the hurricane. And SK Hynix is ​​clear that the worst is yet to come. Guateworse. Different leaders of companies related to AI and components have already commented that we are experiencing an unusual situation, which Years of wild investment in AI remain and that, perhaps, everyone wants to have what they are going to need for the next 10 years as soon as possible, hence the shortage. The heads of the three companies mentioned do not usually speak out, but a few days ago, the CEO of one of the protagonists opened the microphone to warn that things will not improve a little. Kwakk Noh-jung is the CEO of SK Hynix, the other South Korean company that dominates the scene, and a few days ago commented that 2027 will bring the worst memory chip crisis ever experienced. Beyond the depressing forecastcommented that demand is expected to exceed supply beyond 2030, which coincides with some of the pessimistic forecasts from industry heavyweights. goes along the line What other CEOs of related companies have also said, such as Sanjay Mehrotra, CEO of the American company Micron, who in the June results presentation commented that they expect the shortage to persist beyond 2027 due to the very high demand. In addition, it showed that Micron can only satisfy 60% of its customers’ demand. New plants. To try to remedy the situation, the big three are investing in new plants and clean rooms to increase memory production (for AI platforms, of course, not for consumer products). In the case of SK Hynix, as we see in Reutersare considering opening factories in the United States, Japan and Southeast Asia. Kwak notes that “nothing has been decided yet,” arguing that he is evaluating which location can provide the greatest advantage for his business. The two South Korean giants are participating in a government plan to double the country’s memory chip production capacity over the next five years, with $266 billion each to build chip production facilities. In the United States, for its part, similar quantities are also being handled. Micron has changed its investment plan of 200,000 million for one of 250,000 million to address the shortage, with its eyes on the Clay plant in New York, which will become one of the largest semiconductor manufacturing centers in the history of the country. Looking for options. Meanwhile, the crisis continues and companies that create consumer products must find solutions. It has been reported that some like Asus could be interested in opening their own factoriesthis being a slow and very expensive process because creating a clean room from scratch is not easy. Another option is to purchase installations from other companies or look at alternative sources. For example, Apple, after raising the prices of iPad and Mac by an average of 20% (and we will see what happens with the iPhone 18 Pro), would be interested in the memory of the Chinese CXMT. And speaking of China, Huawei is working to expand its empire from semiconductors to RAM with a new plant. Demand on the horizon. All this happens while, in the United States, a demand collective action against Samsung, SK Hynix and Micron for alleged price manipulation and coordinated restriction of DRAM memory production. In the presentations of results and share values, we already see that this crisis has been good for the three companies and the curious thing is that it would not be the first time it has happened. At the beginning of the century, the three companies were investigated for an abusive pricing case that ended with Samsung and SK Hynix fined and Micron getting off the hook because of their cooperation. In China they were investigated again between 2016 and 2018 for a similar case, so we will have to wait and see what happens, but the fact that the price of machines like the Steam Deck has risen almost 300 euros out of nowhere… no one can take it away from us anymore. Buy now. And that is the best advice. Although it is not the one you probably want to read, because you would much rather know when the crisis will end, the best recommendation is that, if you think you are going to need any product soon or need to renew equipment, you better pay the current huge surcharge rather than having to face the one they are expected to have in a few months. Some companies have already mentioned exactly that, that you buy as soon as possible the technology that you think you are going to need for the next five years because things seem like they are going to get much worse before they get better. If it improves, because there are also companies that point out that lPrices will never drop to pre-crisis levels. In Xataka | Memory experts see no relief: “Potential price increases are higher than expected due to shortages”

A new study reveals its crucial role in memory consolidation

by

The nap is an appointment that many people have during the afternoon and that is very difficult to leave aside. For decades, this practice has oscillated between being a purely Mediterranean cultural trait and a productivity “hack” repeated ad nauseum. However, taking a short break during the day is not a luxury for some, but is a top-level brain maintenance mechanism. They have studied it. To prove exactly why napping is so beneficial, a study chose to investigate the effects it has on the brain. Their objective was none other than to check if a short period of sleep could trigger the same ‘brain cleansing’ processes that occur during sleep. And the answer is a resounding yes. Neuronal “saturation”. To understand the impact of the study, you must first understand the synaptic homeostasis hypothesis. This tells us that since we wake up, our brain processes information non-stop and each stimulus and each new piece of information strengthens the connections between our neurons. A classic example may be in a student who is trying to learn an opposition syllabus. The problem is that this continuous increase in synaptic strength consumes a lot of energy and takes up physical and metabolic space. The really important thing here is that there comes a point in the day when the brain is “saturated” since cortical excitability is so high that the ability to consolidate new information plummets. This is when the system literally asks us to “reset” in order to continue processing the information to which we are exposing it. How have they seen it? The experiment involved 20 young adults, and rather than relying on subjective surveys of “how rested they felt,” the team turned to direct physical measurements. They used Transcranial Magnetic Stimulation to measure corticospinal excitability and electroencephalograms to monitor brain activity. Once the method was in place, it was decided to evaluate the patients first between 1:15 p.m. and 2:15 p.m., after spending a good amount of time awake, and then after taking a 45-minute nap. The results. They clearly saw with these analyzes that the brain cleaned its neurons and irrelevant connections were weakened, lowering the “background noise” and returning the system to an optimal state to create new connections again. But also, by releasing this charge, the neurons recovered a very high capacity to induce Long Term Enhancement. That is, the brain was once again in optimal conditions to create a lasting memory. The 20 minute rule. A mantra that can be heard is that the optimal nap should be twenty minutes long in order to regain alertness quickly. However, this study indicates that, to achieve a true architectural recalibration of plasticity in the cerebral cortex, entering into a cycle of about 45 minutes allows the memory consolidation mechanisms to act in depth. In this way, taking a break during the day is not at all something ‘lazy’ but rather it is a brain recalibration system that will allow us to increase our productivity when working or studying. In Xataka | In Spain we have glorified the long nap. In scientific studies they have a different opinion on the matter.

South Korea has a plan to dominate in memory chips and robotics. One of a billion dollars

by

South Korea has put on the table a megaproject for the AI ​​era. This is an initiative made up of three public-private projects spread across semiconductors, data centers and the promising industry of “physical AI”, that is, robots and autonomous systems. The advertisement seeks something very specific: that the country does not depend only on selling memory, but on manufacturing other physical products that it anticipates will be consumed massively. Memory chips rule. The largest item of this ambitious plan is, as we all expected, the one destined for the country’s semiconductor industry. Samsung and SK Hynix have committed to investing $585 billion in new manufacturing plants in the southeast of the country, in addition to strengthening factory construction in the Seoul region. They want to double the production of DRAM memories in five years. The future belongs to data centers. The second large part of this plan corresponds to the data centers that are precisely going to take advantage of a large part of those memory chips. SK Group, CS Group and Naver will invest $357 billion to build large-scale AI data centers in areas until now far from the country’s technology centers. According to the Ministry of Science, the final ambition is to achieve a combined capacity of 18.4 GW by 2035, which would make South Korea one of the world’s great AI “nodes.” Robotics, of course. More surprising is the third leg of the plan: the South Korean government has declared that “physical AI” is a “national strategic industry.” These systems, which allow robots or autonomous vehicles to interact with the real world, also want to be part of the future of the country’s technology industry. In three years they want to create a foundational AI model with the philosophy of “models of the world“—the same in which Yan LeCun or Fei-Fei Li work—. Hyundai has its own plan. The South Korean auto industry giant has set aside $5.8 billion to create a robot manufacturing plant and data center. It’s no coincidence: Hyundai is in fact the parent company of Boston Dynamics since 20221, and is already using its local supply chain to help the American company increase production of its Atlas humanoid robot. The goal: build 30,000 units per year in 2028. Humanoid robots in factories. South Korea wants to commercialize humanoid robots in several key industries before 2028, and to achieve this it wants to train the people who will work with them. The joint plan includes a section that talks about training 10,000 new “AI robotics specialists” in the next five years. But. The announcement coincides with an important internal debate: there are political proposals that seek distribute part of the extraordinary profits of the chip manufacturers, and the unions already see the imminent threat of robotics that will replace positions in all types of assembly chains. The opposition to the South Korean government has also criticized the location of these new production centers, which according to them respond more to a political calculation than to a good industrial strategy. In Xataka | Samsung had been the absolute king of technology in South Korea for decades: SK Hynix has just surpassed it

Apple has a plan for its memory problem. The only problem is that it goes through a company on the US blacklist, according to FT

by

The perfect storm hits the memory market and shakes consumer electronics like we have rarely seen. This stretch of the year has become a bad time to update your computerbuy a console or even expand the RAM of a gaming desktop with a new module. The effects They also reach storage units and other components that until not so long ago seemed more or less predictable. Looking back, that problem with graphics cards during the rise of cryptocurrency mining It begins to seem, for many users, almost small next to what we are seeing now. For months, some large manufacturers have tried to buy time with a tool that the user does not see: inventory. Lenovo is a good example. According to statements by its financial director, Winston Cheng, collected by Bloomberg TV and later cited by various mediathe company even had stocks of memory and other critical components around 50% higher than usual to cushion the blow of shortages and price increases. This type of mattress helps to resist better than others, but it also has a limit when the market continues to tighten. Memory has become a problem for Apple, prices and geopolitics And that’s where Apple comes in. We are not talking about a small company trapped by a specific increase in costsbut one of the companies with the most negotiating power in the entire technology industry. Its supply chain has been operating with extraordinary precision for years, supported by huge agreements, global suppliers and a purchasing power that few can match. Precisely for this reason it is so significant that, according to Financial TimesApple would also be looking new ways to relieve memory pressure. The path that has come to light is especially delicate. According to the British business newspaper, Apple is pressuring the Trump Administration to obtain authorization, or at least a favorable signal, that allows it to buy memory chips from CXMT, a Chinese manufacturer singled out by the Pentagon for his alleged links with the People’s Liberation Army. The media adds that the company approached the Department of Commerce more than a month ago and has also sought support in other parts of Washington. The goal would be to alleviate the financial pressure caused by the rising cost of memory. If you’ve been following Apple for a while, you probably already know: these types of moves are almost never made public. The company does not usually explain its internal purchasing operations, much less its conversations with suppliers or administrations. Financial Times attributes the information to six people familiar with the mattera relevant basis to take it seriously, but we are not facing official confirmation from Apple, the White House or the Department of Commerce. To understand the dimension of the movement, CXMT must be well located. We are not talking about a brand of consumer RAM that the user chooses in a store, but about ChangXin Memory Technologiesa Chinese DRAM manufacturer founded in 2016. The company presents itself as a supplier of chips for mobile phones, PCs, tablets, servers and other equipment, memory that can then end up integrated into finished products. In the case of Apple, therefore, the debate is not about a visible brand, but about who supplies a particularly sensitive internal part. The search for alternatives does not come out of nowhere. It’s no secret that the memory industry is highly concentrated on Micron, Samsung and SK Hynix, the DRAM suppliers that Apple relies on for its devices. When the market is balanced, that dependence can be managed with contracts, volume and planning. When prices rise sharply, however, each additional supplier counts. CXMT appears there as a possible way to add capacity. The memory industry is highly concentrated in Micron, Samsung and SK Hynix. As we have seen, what is behind it is not just an isolated increase in prices, but a change of priorities. The AI ​​craze has put advanced memory at the center of the data center business, especially HBM, needed to power high-performance servers and accelerators. That demand has contributed to a prolonged shortage of traditional memory for consumer electronics. That is why the problem is felt in very different products: not because they all use the same memory, but because they compete, directly or indirectly, for a limited industrial capacity. At Apple, that pressure has ended up reaching the showcase. The company raised the prices of MacBooks and iPads by around 20% and attributed the movement to some “unsustainable” memory prices. The same medium maintains that the decision had an immediate stock market impact: $263 billion less in capitalization, Apple’s second largest daily drop. It is advisable to read these figures together, because they tell something broader than a specific increase: when memory triggers, even a company used to protecting its margins begins to transfer tension to the final product. The expression “blacklist” helps to understand the seriousness of the matter, but it can lead to a conclusion that is too quick. CXMT is on the Pentagon’s 1260H list for its alleged links with the People’s Liberation Army, a very sensitive label in the midst of the technological rivalry between the US and China. However, according to the Financial Times, that designation does not automatically prevent Apple from purchasing chips from CXMT or YMTC. The issue is what would happen if the US hardens its position later. The British economic newspaper says that the Department of Commerce had already included CXMT in a package of Chinese companies candidates to enter the Entity List, a much more restrictive trade listbut that the White House asked to wait for negotiations with Beijing. That background explains why the company would want a political signal before moving. It is not enough for a door to be open today if tomorrow it can be closed with Apple inside. John Moolenaar, Republican chairman of the House China committee, told the same newspaper that partnering with a Chinese military company would be “a serious mistake.” He also argued that helping the “Chinese … Read more

We thought we had passed the worst of the memory crisis. We were totally wrong

by

A new report from the consulting firm Jefferies Equity Research paints an even more terrible picture than we expected for the immediate future of the memory crisis. These components have already risen in price extraordinarily in the last nine months, but wait: it hasn’t ended there by any means. The rest of 2026 will be horrible. According to its analysts, memory prices will experience an increase of between 40 and 50% in the third quarter of 2026 compared to the current quarter. But the increases will not stop there, because in the fourth quarter, between 30 and 40% price increases are expected compared to the third quarter. Let’s do the math. According to this prediction, the prices of memory modules for our PCs will be absolutely exorbitant: 16 GB DDR4 module: the current price is around 139 euros. Applying these increases, in the third quarter this product will cost up to 209 euros in the third quarter, and up to 292 euros in the room. 16 GB DDR5 module: the current price is around 240 euros. Applying the same percentages, in the third quarter we would pay up to 360 euros for it, and in the fourth quarter up to 504 euros. Bad for (almost) everyone. These increases will therefore make prices much worse than they are now, and although we have given the example of memory modules for the PC, this problem extends to all types of electronic products that have this component. Mobile phones, tablets, graphics cards, Smart TVs, routers, consoles (Hello Steam Machine) or cars will also be affected, and this could therefore significantly impact the future prices of these products. Let them tell Apple: we could see how this and other companies are forced to raise prices again. 2027 will be bad too. This report also reveals that the crisis will persist in 2027, although the growth curve will reduce slightly: increases of between 40 and 45% are expected from year to year. It is a very bad figure, but not as bad as what is expected for this second half of 2026. Why does the crisis last so long?. AI is to blame for this memory crisis, and what is happening is that manufacturers are not only unable to cope: the “little memory” they are manufacturing is being reserved for long-term contracts. Micron, for example, has already signed 16 strategic contracts with large firms and hyperscalers, and in the global market 50% of what is manufactured is already reserved for those large clients. That percentage could rise to 70%, they say at Jefferies, which will make it even more complicated (or rather, more expensive) to access memory modules for PCs, laptops, consoles or mobile phones. The end of the crisis, in 2028? This analysis also indicates that 2028 could finally see prices begin to fall slightly. The reason: a slowdown in demand and a slight increase in supply of between 15 and 20%. China manufacturers won’t save us. He Chinese manufacturer CXMT it seemed the great hope for consumers, but as seen at Computex, their prices they are similar to those of Micron, Samsung or SK Hynix. Apple is trying to work with that Chinese firm and is pushing for the US I took it out from your “Entity List”, but it remains to be seen what will happen. The only advantage of this and other Chinese manufacturers such as YMTC is that they do have modules in their inventory, but most are intended for domestic consumption by Chinese firms. In Xataka | Samsung had been the absolute king of technology in South Korea for decades: SK Hynix has just surpassed it

An unexpected salvation for the end user emerges from the memory market debacle: Chinese chips

by

The DRAM memory industry is facing a profound structural transformation. Until October 2025 the price of memory chips evolved in a relatively stable way, but from that moment on began a dizzying climb which still continues. In fact, the consultant TrendForce expects the price of conventional DRAM to rise between 58% and 63% quarter-on-quarter before the expiration of the second quarter of 2026. And the artificial intelligence (AI) is behind all this. The three largest chip manufacturers of memory on the planet, the South Korean companies SK Hynix and Samsung Electronics, and the American Micron Technology, They have reallocated about 70% of its production lines to high-bandwidth memories (HBM) to satisfy the currently insatiable demand of data centers specialized in AI. The consequences of this movement did not take long to appear: standard DDR4 and DDR5 memories and their derivatives, which are the most used in the consumer segment, immediately began to become scarce. And its price skyrocketed. In fact, according to the consulting firm GartnerRAM has gone from representing 16% of the total cost of a laptop to 23%. And it is possible that this escalation will continue to develop in the coming months. However, users can cling to the greatest stabilizing agent in the memory market today: Chinese manufacturers. This is the great opportunity for YMTC and CXMT Yangtze Memory Technologies Co. (YMTC) is one of the largest NAND chip manufacturers in China. Its global market share is approximately 13%making it one of the main competitors of Samsung, SK Hynix, Micron, Kioxia or SanDisk. Its weight in the Chinese market is very great, especially because US sanctions They prevent American and South Korean memory manufacturers from selling their most sophisticated integrated circuits to their Chinese customers. On the other hand, Changxin Memory Technologies (CXMT) is one of the Chinese companies specialized in the production of memory chips, and, like other companies in the country led by Xi Jinping, it has chosen to compete in this very attractive market by deploying a very aggressive pricing policy. CXMT in particular has increased its DRAM chip production capacity almost five times during the last four years, which has allowed it to increase its global market share until reaching a very worthy 7.6%. CXMT has chosen to compete in this very attractive market by deploying a very aggressive pricing policy. While large foreign manufacturers maximize their margins thanks to data centers and the rise of AI, Chinese manufacturers prioritize sourcing from local companies. This scenario allows the supply and prices of memory and NAND chips in China to remain relatively stable, remaining outside the strong premiums charged by the big three (Samsung, Micron and SK Hynix). This is the context in which the Chinese memory module brands Gloway and KingBank have recently announced new DDR5 modules that stand out for using SDRAM memory chips made in China. With a standard configuration of eight chips per module, these companies can produce 24 GB modules and group them into kits of two or four modules to achieve capacities of 48 GB or 96 GB, respectively. Chinese memory chips, particularly those from CXMT, have already begun to spread beyond China’s borders. Corsair has already integrated them into some kits of its Vengeance line, while HP and Dell have begun the process of homologating modules with CXMT chips for their products. This is good news for users, there is no doubt. Even so, we still don’t know if the use of CXMT DRAM will become widespread in response to AI-induced shortages. The market demands new players, wherever they come from, and if YMTC and CXMT are able to fill the gaps left by Samsung, Micron and SK Hynix, they are welcome. Image | Intel More information | Tom’s Hardware In Xataka | China needs to develop a new type of chips immune to US sanctions. And your scientists have just achieved it

What is memory in AI and how to make the most of it and use it correctly

by

Let’s explain to you what is AI memory and how to use it well to get the most out of it. This is a feature that is already included in practically all the main chatbots with artificial intelligence, and it helps you give them context about you and apply it in their responses. If you often use bots like ChatGPT, Gemini or Claude, you may have noticed that there are things that are not remembered from one chat to the next. Or it may be the other way around, you may be surprised that he remembers a specific detail that you didn’t expect him to remember. This is precisely due to the way in which the two types of memory that AI has work. What is AI memory When we talk about memory in artificial intelligence, we are referring to an internal system that these chatbots have to save information about you between conversations. Not only within the conversations themselves where the context is maintained, but also between different conversations. For example, ChatGPT, Gemini or Claude may remember details such as the city where you live, your profession, or even your personal tastes. They can also remember if you have pets, what pets you have, your projects, the topics you usually write about, your style, and many other preferences. This memory is used to contextualize the answers that the AI ​​gives you based on your data that may make you prefer one type of response or another. So, even if you start a conversation from scratch in each chat, there will be some specific information that continues to be remembered so that you don’t have to write it over and over again. What types of memory does AI have? Artificial intelligence does not treat all the memories it has about you the same, and we can distinguish between two different types of memory. Each of them has its specific use. They are the following: Saved or explicit memory: This memory is like a kind of personal file about you that the AI ​​generates by saving specific data about you. It can do this automatically when you mention something from your context, like your job, but you can also ask it to remember something about you with a command like “remember that I’m a vegetarian.” History-based memory: Instead of saving specific data, some AIs can also consult the log of your previous conversations. This does not generate a profile of you, but rather they analyze the content of everything you have said before to give you more coherent answers. In practice, the difference is that the saved memory is that in which the AI ​​saves the elements that you ask for or those that it considers appropriate, and from them it generates a file about you that it updates. So, you then use that knowledge to generate the answer. Meanwhile, history-based memory is as if the AI ​​could reread its notes from previous conversations before responding to you. If you don’t want the AI ​​to remember a conversationAI systems also have an incognito mode, like in browsers. With it, chatbots will neither save data about you that you write in it nor will they remember the chat in the future. This conversation will not be saved in your history either, although the data you write does end up on the company’s servers. Why AI “forgets” by default An artificial intelligence language model like the one behind ChatGPT, Claude or Gemini, only processes what is within the active conversation. When you ask something, it will take into account everything you have asked it during this conversation, and once you close the conversation or open a new one, the previous one is forgotten or not taken into account. This is not a mistake, but a deliberate decision. On the one hand, it allows you to maintain the context of everything you are asking within the same conversation, while in each new one you can start from scratch without taking into account everything written in another chat. Beyond the practical reasons there are also the technical ones. Maintaining the context of millions of simultaneous chats permanently would require a large consumption of data and energy. Additionally, in the event of a security breach, if the AI ​​has already forgotten past conversations there will be fewer risks to users’ privacy. AI chatbots usually allow you to manage memory that they have stored. On the one hand, you will be able to edit or delete the explicit memory with the file they have made about your data, and on the other hand, you will also be able to configure how the history memory is managed. There is an important detail that should be remembered: Deleting a chat does not delete the saved memories that were created in it. They are two different types of memory. If you delete a conversation in which you have told ChatGPT that your hair is blue, the AI ​​will not refer to what you have talked about with it and will have forgotten it, but if it has saved in its explicit memory that your hair is blue, this is something it will always know in all the chats you start with it. Additionally, many AIs have a system to import memory from other competing AIs. so that you can pass from ChatGPT to Claude either from Claude to Gemini taking you from one to another explicit memory with everything an AI knows. How to make the most of AI memory AI memory can save you a lot of time when you use it well, but if you leave it to its own devices it can also accumulate incorrect, outdated or useless information, and cause frustration or incorrect answers. For example, I once asked ChatGPT to add a cumquat to an image, and until I removed it from memory, for months it was adding cumquats to all the images I asked it to. Therefore, it is important that periodically review … Read more

What science knows about how a hazelnut affects your memory

by

A nutritional recommendation that we have been told for a long time is that every day we should eat a handful of nuts to be able to maintain good health in the long term. Here, experts specifically point out that it is not enough to eat walnuts or almonds separately, but the key is to consume mixed nuts to prevent cognitive decline. More and more voices. This is what genetic epidemiologist and microbiome expert Tim Spector has shown, who has pointed categorically that “the best way to help your brain health is to eat mixed nuts.” The problem is that, from a scientific point of view, although there is evidence that reinforces this idea, there are also others who do not think the same and do not see that nuts are an infallible prevention against neurodegeneration. It is studied. To understand how a pistachio or hazelnut can affect our memory, we must first look at the intestine. Here, a comprehensive systematic review published in 2024 in Signal Transduction and Targeted Therapy details how the intestinal microbiota It not only digests food, but regulates the functions of glial cells in the brain, becoming a therapeutic target for neurodegenerative diseases. And it is no wonder, because glial cells act as the defense system that our brain has. How it works. The mechanism proposed here mainly focuses on the theory that consuming the fiber and polyphenols present in nuts causes our intestinal bacteria to ferment them and produce short-chain fatty acids. These metabolites are capable of passing directly to our brain, reducing neuroinflammation and promoting neuronal plasticity. In fact, recent analyzes of more than 500 studies confirm that a healthy microbiota is directly associated with better connectivity in areas of the brain dedicated to memory, attention and emotions, such as the amygdala and frontal cortex. In the long term. The clearest evidence we have to date connecting nut consumption, microbiota, and cognition comes from Spain. Here is a prospective study published in the journal Age and Aging by researchers from the University of Rovira and Virgili, where it was analyzed to 747 overweight patients with metabolic syndrome for six years. Based on this sample, they began to give different portions of 30 grams of nuts to the patients and waited to see what would happen. The results suggested that with 3-7 servings of nuts a week, a significantly slower cognitive decline can be maintained. Beyond. The most interesting thing was the “how” since when analyzing the participants’ feces, they saw that the consumption of nuts increased the Shannon index, which is a marker of how rich our microbial life is in the intestine. In this way, it was seen that the consumption of nuts caused a increase in bacteria that are very beneficial for us, such as Lachnospiraceae UCG-004whose abundance was directly linked to better mental performance. Why ‘mixed’. The insistence on a variety of nuts has a microbiological basis, since systematic reviews show that each type of nut feeds slightly different bacterial families. In this way, while almonds can favor certain strains of bacteria after four weeks of consumption, walnuts, hazelnuts or pistachios enhance different ones. That is why, if we feed all the strains, diversity will be maximized, which is what interests us to generate a greater variety of neuroprotective metabolites. It’s not magic. At the moment, all the studies we have on the table are observational, so they demonstrate a strong association between eating nuts and cognitive improvement, but it is not absolute causality. Here, a meta-analysis of 2024 published in Frontiers in Nutritionwhich reviewed 5 clinical trials with 928 adults, did not find a significant overall effect of nut consumption on cognition in the short or medium term. But more recent individual trials also show some doubtssince, while a dose of 50 grams per day of almonds failed to improve long-term memory in prediabetic adults according to a study in Nutrition Journal, another similar test in the Asian Indian population did report improvements in executive function after 24 weeks. Images | Pratik Bachhav In Xataka | The pistachio has worked an unexpected wonder: generating thousands of jobs in the fields of Castilla-La Mancha

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.