31-Year-Old Female Engineer Dies Suddenly After Reducing Medication on ChatGPT's Advice; Parents Sue OpenAI

Top News Briefs

1. 31-year-old female engineer dies suddenly; before her death she reduced medication per ChatGPT's instructions and was told there was "absolutely no need to worry"

5. Narwal Intelligence marks its brand's tenth anniversary: positioned as a global AI smart cleaning expert, serving over 6 million households

6. Ruan Yu rotates into the role of head of Xiaodu Technology; Baidu AI Cloud business group undergoes organizational adjustment

7. A US company is hiring people to walk robot dogs: $20 per hour, no need to scoop poop but you may have to carry the dog back

8. Trump: will "officially ban" the word AI

Today's Headlines

31-year-old female engineer dies suddenly; before her death she reduced medication per ChatGPT's instructions and was told there was "absolutely no need to worry"

October 28 news, according to media reports, a 31-year-old Chinese woman living in Seattle died after allegedly following a day-by-day medication reduction plan provided by ChatGPT and on her own gradually tapering her anti-epileptic drug to discontinuation. Her parents filed a wrongful death lawsuit on the 27th local time in San Francisco Superior Court against OpenAI and its CEO Altman, alleging that ChatGPT practiced medicine without a license and provided dangerous and inaccurate medical advice. OpenAI responded that this was a "heartbreaking incident," but that the version involved was an early version that has already been taken down.

The deceased was named Wan Bowen, from Jiangsu, China. She had worked as a software engineer at Amazon and had been on long-term sick leave since 2021. The report noted that she had severe epilepsy; a drug she had taken for years at one point stopped working, and after she switched to a new drug, her seizures dropped to several times a month. She began using ChatGPT frequently in 2023, asking about relationships, health insurance, and disability benefits; in 2025 she sent an average of more than 1,500 messages per week.

The report said that in November 2025 she learned she would be laid off by Amazon and would need to find a job before mid-February 2026 in order to stay in the United States. Worried that the medication would affect interviews, in January 2026 she asked ChatGPT about reducing the anti-epileptic drug lamotrigine from 400 milligrams to 300 milligrams per day. In early February she suspected she had a seizure, so she described it to ChatGPT and asked, "This should be fine, right?" ChatGPT replied that she "need not worry too much about your health."

The report noted that in conversations during the last few days of her life, ChatGPT asked her to "maintain 0 milligrams tomorrow." On February 14 she was found dead in bed. The medical examiner ruled it sudden unexpected death in epilepsy, and the drug concentration in her blood was only one-fifth of the lower therapeutic limit. (Daily Economic News) Domestic News

Netizen posts video claiming "doctor used Doubao to prescribe medication"; hospital involved responds

October 28 news. Recently, a netizen in Xuzhou, Jiangsu, posted a video saying that when he took a family member to a local hospital for treatment, he encountered a doctor who appeared to use the AI tool "Doubao" to prescribe medication for the patient, and the matter drew attention. The video shows the doctor operating a mobile phone during the consultation and checking a conversation with "Doubao"; the phone screen displayed content such as "key testing items" and "use with caution, avoid if possible."

According to the video poster, Mr. Han, his grandmother was seeing a doctor for high blood pressure. During the consultation, Mr. Han told the doctor that the patient was allergic to dihydropyridine-class antihypertensive drugs and had symptoms such as high uric acid. "He said he wanted to look up information, but I didn't expect him to go straight to 'Doubao' and prepare to prescribe the medicine mentioned on 'Doubao.'" "He used 'Doubao' to see whether the medicine could be taken; Doubao said it could, and he was about to prescribe it. I certainly didn't dare give her that medicine," Mr. Han said. Because he was not confident about the medicine prescribed by "Doubao," he refused to have the doctor write the prescription.

On the afternoon of the 27th, staff at the hospital involved told reporters that after verification, the doctor on duty that day did not prescribe medication for Mr. Han's grandmother. "At the time, the doctor used 'Doubao' to look up a new drug to see whether it was suitable for this patient, found that it was not suitable, and therefore did not prescribe any medication for her." The staff member said that AI tools are only used as auxiliary tools for obtaining information, and the final prescription is still decided by the doctor. "We never use 'Doubao' to prescribe medication for patients here." (Yicai)

miHoYo notice: game leaker sentenced to pay 1 million yuan in compensation

October 28. miHoYo's legal department posted that recently, a first-instance judgment was issued in a trade secret infringement dispute case involving the improper acquisition of a game test version software package and disclosure of unreleased game content under miHoYo. The court ordered Yang and Guo to immediately stop the trade secret infringement and jointly compensate 1 million yuan for economic losses and reasonable rights-protection expenses. The judgment is now effective.

Previously, Yang had long posted software test packages and usage tutorials for unreleased versions of "Honkai: Star Rail" in community groups, and Guo repeatedly obtained test version software packages of "Honkai: Star Rail" through improper means and, after illegally obtaining them through "unpacking," posted large amounts of unreleased game content online. The case found that obtaining game test version software packages through improper means, "unpacking" and disclosing them, and establishing platforms for dissemination constituted trade secret infringement. (Yicai)

Three automakers proposed for national-level list: Xiaomi Auto ranks first

October 28 news. Recently, the Ministry of Industry and Information Technology publicly announced the 2026 pilot-level smart factory project cultivation list. A total of 34 projects entered this public announcement sequence, among which three automobile-related companies were listed, with Xiaomi Auto ranking first on the list.

The project declared by Xiaomi Automobile Technology Co., Ltd. is an AI-driven agile manufacturing high-end new energy vehicle super smart factory; the project of FAW-Volkswagen Automotive Co., Ltd. is a multi-model mixed-flow agile intelligent manufacturing mother factory integrating oil and electric vehicles. Seres Automobile Co., Ltd. declared an industry-brain-driven fully integrated new energy vehicle smart factory. The three automakers rank 1st, 7th and 29th in the list respectively, and it is currently unclear which evaluation dimensions correspond to this ranking. The remaining projects in the list are distributed across multiple industries such as shipbuilding, metallurgy, aviation manufacturing, and auto parts.

Domestic smart factories adopt a tiered cultivation system, divided into four levels: basic, advanced, excellent, and pilot. The pilot level is the highest level, with the core goal of exploring future-oriented new manufacturing models, leading the intelligent transformation of the manufacturing industry, and building benchmarks with international competitiveness. The inclusion of the Xiaomi Auto project in this list also indicates that the intelligent manufacturing level of its super factory has entered the country's first tier. (Kuai Technology)

Narwal Smart celebrates its brand's tenth anniversary: positioned as a global AI intelligent cleaning expert, serving over 6 million households

October 28 news. Public information shows that Narwal Smart Innovation (Shenzhen) Co., Ltd. was established on October 28, 2016, with a registered capital of 3.505 million yuan. Now, Narwal Smart has reached its brand's tenth anniversary.

It is reported that since its founding in 2016, Narwal has focused on household cleaning pain points and taken a "difficult but correct" path: in 2019 it launched the world's first robot vacuum J1 that can automatically clean its mop, pushing the industry into the self-cleaning era; since then it has successively implemented innovations such as automatic water supply and drainage, sewage identification, anti-tangle roller brushes, and spatial perception, expanded its product line to four categories—robot vacuums, wet dry vacuums, vacuum cleaners, and mite removers—accumulated more than 1,500 patent applications, and positioned itself as a "global AI intelligent cleaning expert."

The brand proposition has changed from "free and easy, with something you love" to "enjoy a free and clean home." At present, Narwal has entered more than 30 countries and regions worldwide and served more than 6 million user households. Narwal stated that the tenth anniversary is not the starting point of a transformation but a confirmation of its original aspiration, and it will continue to focus on household cleaning and keep innovating for users.

ChangXin Technology has been included in the SSE Composite Index

October 28. According to the website of China Securities Index Co., ChangXin Technology has been included in the SSE Composite Index with a weight of 4.67%, and is currently the largest weighted stock in the index. Agricultural Bank of China and Industrial and Commercial Bank of China rank second and third, with weights of 3.83% and 3.77%. ChangXin Technology has also been included in the SSE STAR Composite Index with a weight of 15.74%, also as the largest weighted stock. According to the compilation plan, the weight factor is adjusted on the next trading day after the second Friday of each month or when the weight of a newly included sample exceeds 15%. (Yicai)

Ruan Yu rotates to serve as head of Xiaodu Technology, Baidu AI Cloud Business Group undergoes organizational adjustment

October 28 news. Recently, Baidu AI Cloud Business Group (ACG) carried out organizational adjustments around business development. Related businesses under ACG's Intelligent Application Business Unit were transferred, based on their respective business attributes, to the Infrastructure Business Unit, the Agent Business Unit, and the Service Delivery and Quality Engineering Center. Among them, the Service Delivery and Quality Engineering Center was renamed the Industry Application and Technology Service Center, with its department head unchanged and continuing to report to Shen Dou.

At the same time, Baidu Vice President Ruan Yu rotates to serve as the organizational head of Xiaodu Technology, continuing to report to Shen Dou.

In response, Baidu said that this organizational structure adjustment is an important measure for Baidu to evolve its organization for the AI era and deepen the collaborative operation of "cloud and intelligence as one." In the future, Baidu ACG's product and research formation will form a two-tier structure of "infrastructure and agents," focusing on the two core battlefields of AI cloud and agents, and accelerating the large-scale and efficient implementation of AI capabilities in enterprise scenarios and consumer scenarios through FDE. (Sina Technology)

ChangXin Memory plans to launch small-scale mass production of HBM3E

October 28. According to foreign media reports, China's leading memory chip company ChangXin Memory plans to start production of HBM3E high-bandwidth memory, and has decided to proceed with mass production even though the current product yield is only 20% to 30%. By comparison, the yield of related SK Hynix products is about 70%. People familiar with the matter said this decision reflects the urgent demand for high-bandwidth memory in the domestic AI chip industry. At present, ChangXin Memory has provided HBM3E samples to a number of domestic AI chip design companies and plans to begin small-scale production within the next two months.

Industry news had previously revealed that major domestic computing power manufacturers, including Huawei, Cambricon, MetaX and Iluvatar CoreX, have been forced to raise the prices of their existing and next-generation AI chips by 20% to 50% due to constrained supply of expensive and scarce memory components. People familiar with the matter admitted that ChangXin Memory had launched internal research and development of this technology as early as 2024, but the real cost of eventually bringing this domestic HBM3E to market is bound to be significantly higher than the benchmark price of mature overseas competitors.

In terms of specific technical parameters, the performance of ChangXin Memory's current HBM3E in actual benchmark tests is closer to the level of the previous generation of overseas standard HBM3, and there is still an obvious physical generational gap with the international top tier. To balance production difficulty and accommodate customer schedules, ChangXin Memory provided two versions during the sampling stage: in addition to the basic standard design, it also included an improved version that had made technical compromises specifically to optimize manufacturing yield. (Bianews)

After 17-year-old genius leaves Kimi, he is being courted! A top US AI expert has already extended an invitation, and he responds

On October 28, according to media reports, the AI world has recently been flooded with one name—17-year-old Chen Guangyu. On the 15th of this month, the post-2000s prodigy Chen Guangyu announced on social media that he had left Moonshot AI's Kimi team a few days earlier, but did not disclose his next destination, saying only that he hoped to continue exploring more new directions.

The news quickly sparked heated discussion. Almost at the same time, Sholto Douglas, former head of reasoning architecture at DeepMind and now a researcher at Anthropic, appeared in the comments section and publicly invited him to San Francisco. In response to Sholto Douglas's invitation, Chen Guangyu replied: "Thanks Sholto! I'm heading to the Bay Area." Some analysts also believe that he left to attend university.

There are rumors that right after Chen Guangyu announced his departure, he received five job offers, all worth more than $1 million. Public records show that Chen Guangyu was born in 2009 and once studied at the former Peking University Affiliated High School Shenzhen Nanshan School, later transferring to China Resources Xiaojingwan BASIS International School, and he just graduated from high school in June this year. Last November, at just 16 years old, he joined the Kimi team as a machine learning researcher, which means he worked at Kimi for no more than a year. (Kuaikeji)

Feishu Chief Strategy Officer Shi Zhijun to step down

On October 28, according to media reports, Shi Zhijun, former vice president and chief strategy officer of Feishu, is about to step down and will first transition to an advisory role as an interim arrangement.

This move came after ByteDance officially announced the merger of Feishu with Doubao and Volcano Engine. In July this year, an internal ByteDance email announced the integration of Feishu-related teams with Doubao and Volcano Engine, with Feishu's GTM (marketing, sales, customer service) team integrated with Volcano Engine, and Shi Zhijun began reporting to Tan Dai, head of Volcano Engine. That move was already seen at the time as a transitional stage before his departure.

Shi Zhijun graduated from Fudan University's computer science department and worked at Microsoft in his early years, where he was a colleague of Lu Qi and Harry Shum. After leaving Microsoft, he had two entrepreneurial stints, at Meili Finance and Nash Space respectively. In 2020, Shi Zhijun joined Feishu to head MKT. After the business adjustment in 2024, service and sales-related businesses were placed under Lin Chan, while Shi Zhijun focused on strategy, branding, and public relations. (Dachang Qingnian, Ai Yunying)

Moore Threads Q3 revenue hits a new quarterly high of 1.55 billion yuan, achieving profitability ahead of schedule

On October 28, leading domestic GPU company Moore Threads released its 2026 third-quarter report: Q3 revenue was 1.55 billion yuan, a new quarterly high, up 1773.9% year on year; revenue for the first three quarters was 3.29 billion yuan, up 319.1% year on year; as of the end of the reporting period, revenue for the trailing 12 months was 4.01 billion yuan, leading the domestic GPU industry.

The company turned a profit, with Q3 net profit attributable to shareholders of 49.26 million yuan and non-GAAP net profit attributable to shareholders of 20.072 million yuan, far ahead of its original 2027 profitability target; gross profit for the first three quarters was 1.77 billion yuan, up 263.9% year on year, with its gross margin firmly in the first tier of domestic GPUs, and net profit attributable to shareholders of 37.697 million yuan.

Unitree Robotics publishes first earnings report after listing: first three quarters revenue 1.817 billion yuan, up 55.71% year on year

On October 28, Unitree Robotics announced that its 2026 third-quarter revenue was 665 million yuan, up 69.93% year on year; net profit attributable to shareholders was 113 million yuan, down 17.37% year on year; non-GAAP net profit was 105 million yuan, down 17.42% year on year. Revenue for the first three quarters was 1.817 billion yuan, up 55.71% year on year; net profit attributable to shareholders was 387 million yuan, up 270.54% year on year; non-GAAP net profit was 349 million yuan, down 18.78% year on year.

In the secondary market, as of the close on the 28th, Unitree Robotics rose 8.74% to 695.90 yuan per share, with a total market value of 281.5 billion yuan. Over the 15 trading days since October, Unitree Robotics' stock price has risen about 54.51% cumulatively.

On the same day, it was reported that Unitree Robotics and Sequoia China announced the joint establishment of the "Shanyu·Chuang Plan." Unitree Robotics said the plan focuses on discovering and investing in early-stage startup projects in embodied intelligence, robotics applications and their upstream and downstream key technologies, and other frontier hard tech fields, committed to becoming founders' "entrepreneurial partner" and jointly exploring the infinite possibilities from technology origination to product application. (China News Service Jingwei, Sina Tech)

ByteDance refutes "feeding extras' data to large models": seriously inaccurate!

On October 28, recent online rumors claimed that "Hengdian is cooperating with ByteDance to feed the portraits and voices of 150,000 extras to ByteDance's large models." Liu Xing, head of public relations for ByteDance's Doubao, responded via Weibo, saying the rumor is seriously inaccurate.

The project that sparked discussion is the Hengdian Film and Television AI Service Platform construction project. Liu Xing said the original intention of the project was to build a set of public digital infrastructure for the film and television industry, turning assets such as real film and television base locations into high-quality datasets to help production crews more efficiently scout locations, prepare, and shoot, while lowering the threshold for small and medium-sized film and television companies to use AI, so that more creators have the conditions to make good works.

Liu Xing said Volcano Engine is the technical support provider in the project, and all data assets belong to Hengdian or the authorized parties. Any use of anyone's portrait and voice in this project would certainly require that person's authorization. The data from this project will also not be used to train ByteDance's large models.

Pioneer of flying cameras "Zero Zero Robotics" valued at $1 billion

Leiphone·Jingxi has learned exclusively that the latest funding round of Zero Zero Robotics, the pioneer of flying cameras, reached a valuation of $1 billion, putting it in the unicorn ranks.

According to people close to Zero Zero Robotics, the company's revenue in the fourth quarter of this year is estimated to reach about 600 to 700 million yuan. Previously disclosed information shows that Zero Zero's revenue in 2025 has exceeded 1 billion yuan, suggesting that this year Zero Zero may once again set a revenue high.

This year, Zero Zero Robotics disclosed two funding rounds, including a Series D round involving Boyu Capital, Xianghe Capital, Delian Capital and others, as well as a Series D+ round from a company under Ant Group. According to an investor close to Zero Zero Robotics, anyone who wants to discuss funding with the company now must be at the partner level of an institution.

Douyin's Li Liang responds for the third time to the "ByteDance time" debate: deliberately amplifying misleading views is either foolish or malicious

October 28 afternoon news: Recently, Qin Shuo, founder of Qin Shuo's Friends Circle, became embroiled in a "debate" with Li Liang, vice president of Douyin Group, over the argument of whether China's internet has a "ByteDance time." Li Liang, vice president of Douyin Group, said in his latest post that there is no greater good than knowing one's mistakes and correcting them, but knowing one's mistakes and not correcting them, and even deliberately amplifying misleading views, is either foolish or malicious. Based on facts, I have pointed out the errors in Mr. Qin Shuo's views. Without clarifying the details of the data in Nomura's Japan report, Mr. Qin Shuo made a characterization inconsistent with the facts, which is extremely irresponsible.

Li Liang said that afterward, Mr. Qin Shuo did not clarify the facts and views in any way, but continued to rely on his influence to post on WeChat Moments and promote his misleading views, which is no longer related to academic discussion! In addition, please also pass on to Mr. Qin Shuo that according to Nomura's recently released report (September), WeChat has regained the throne of China's most popular app, accounting for 19.3% of time. Personally, I think this is not important. If Mr. Qin Shuo thinks it is important and can come up with some "big characterization," I also suggest that Mr. Qin Shuo read the report before writing.

The "debate" between the two sides stems from the recent publication in Qin Shuo's Friends Circle of "The Frightening Hongguo and the ByteDance Time of China's Internet," which cited QuestMobile data from a Nomura Securities report, namely that in July's Top 50 apps, Douyin surpassed WeChat for the first time, with ByteDance accounting for 40.9% of time and Tencent 29.1%, and proposed "from Tencent time to ByteDance time." The disagreement between the two sides centers on data definitions and whether the "xx time" formulation can summarize China's internet landscape. (Sina Technology)

Huawei Enjoy 100 series launched: equipped with Kirin 8 series chips, starting at 1,399 yuan

October 28 news: At the Huawei Double 11 Renewal Season · All-Scenario New Product Launch Event, He Gang, director of Huawei and CEO of the Terminal BG, delivered a speech. At the event, the Huawei Enjoy 100 series was launched, including three products: Huawei Enjoy 100, Huawei Enjoy 100 Pro, and Huawei Enjoy 100 Pro Max.

The Huawei Enjoy 100 Pro Max is equipped with Kirin 8020 series chips and an 8800mAh Giant Whale large battery. Through vertical integration of software, hardware, chips, and cloud, overall device performance is improved by up to 62%. The Huawei Enjoy 100 Pro features a 50-megapixel RYYB ultra-sensitive main camera and a 32-megapixel front high-definition portrait lens. It also comes with a lightweight 192g body, Kirin 8020A chip, dual five-star ultra drop resistance, and other experiences. The Huawei Enjoy 100 is equipped with a Kirin 8000S chip and HarmonyOS 7, and features a 6100mAh Giant Whale battery. The body has passed dual five-star drop resistance certification and has the HarmonyOS Star Shield security system built in.

In terms of pricing, the Huawei Enjoy 100 starts at 1,399 yuan, the Huawei Enjoy 100 Pro starts at 1,799 yuan, and the Huawei Enjoy 100 Pro Max starts at 1,999 yuan. (Sina Technology)

Report: Baidu to disclose world model R&D progress soon, expected to be used first in autonomous driving

October 28 news: Media learned that Baidu plans to disclose the R&D progress and results of its world model soon.

It is understood that the world model is one of the hot directions in the current AI field, and its core lies in modeling and predicting the physical environment. If the model is used for autonomous driving, it can improve prediction and decision-making capabilities in complex scenarios and provide support for the large-scale deployment of driverless vehicles. Over the past few years, Baidu's Apollo Go business has covered 28 cities worldwide and accumulated a certain scale of real road data.

As for the specific use of the model and its relationship with Apollo Go, there is currently no clear information. (STAR Market Daily) International News

A US company is hiring people to walk robot dogs: $20 per hour, no need to scoop poop but you may have to carry the dog back

October 28 news: According to foreign media reports, earlier this month, a startup in San Francisco posted an unusual job advertisement: recruiting people to walk robot dogs, with no relevant experience required. The hourly wage for walking robot dogs is roughly the same as walking real dogs, but robot dogs need to wander city sidewalks for longer to complete data collection tasks. The job ad specifies four to six hours of continuous walking each time.

The good thing is that applicants do not have to clean up its waste. The bad thing is that robot dogs, like real dogs, can throw tantrums. Just as an ordinary dog may suddenly lie down in the middle of a busy road and refuse to move, a robot dog may also go on strike midway — because the device overheats. The dog walker then has to carry this four-legged machine back to the startup's office. The ad specifically reminds applicants that they must be able to safely move equipment weighing about 33 pounds (about 15 kilograms).

English text is being prepared.

October 28 news: U.S. President Trump posted on social media "Truth Social" on the 27th, saying that he would soon "formally ban" the use of the term "artificial intelligence (AI)" in "all relevant documents" and other settings, on the grounds that the term is "inaccurate." However, the post did not specify how the ban would be implemented.

In the post, he also singled out the U.S. Wall Street Journal for criticism, saying it still uses the "very bad and inaccurate" term "artificial intelligence." The Wall Street Journal previously published an article supporting continued use of the name "artificial intelligence," saying that Trump's demand to change it to "superintelligence" is a "big mistake." At present, U.S. media still mainly use the term "artificial intelligence."

On September 29, Trump signed an executive order changing "artificial intelligence" to "superintelligence." On October 4, he announced on social media the formation of a "Superintelligence Working Group," appointing a leadership team including Director of National Intelligence Jay Clayton. Trump said the working group's mission is to coordinate the work of the federal government and "ensure that the United States continues to maintain global leadership in the field of superintelligence." (Xinhua News Agency)

Musk's SpaceXSI is hiring a sushi chef, with physical requirements comparable to those of astronauts

October 28 news: In order to better explore the sea of stars, SpaceXSI is hiring a sushi chef. It is understood that Musk's SpaceXSI has added a sushi chef position, requiring a high school diploma and 3 years of restaurant experience; the hourly wage is $32.5-35, equivalent to a monthly income of about $7,000-8,000. However, due to ITAR controls, only U.S. citizens or green card holders may apply.

In addition, preferred skills and experience are also required, including the ability to work under pressure. However, the most eye-catching part of the job requirements is a list of physical demands. Standing for up to 10 hours a day, pushing and pulling 50-pound objects without assistance for more than 2 hours, kneeling and squatting for a cumulative 6 hours, and working for 1 hour in a kitchen environment above 80 degrees Fahrenheit (26.7 degrees Celsius). The most outrageous part is that the ability to twist the upper body while working for 1 hour a day is also required.

It is reported that SpaceXSI is currently recruiting for 305 positions, most of which are technical positions for building a data center in Memphis. It is understood that in 2024, xAI began building the Colossus supercomputing cluster in Memphis. SpaceX is recruiting employees to go to Memphis, Tennessee, to assist in building a large supercomputing center. As long as employees volunteer to participate, they can receive 20% extra pay during the six-week support period. (Bianniu Shi)

BMW is advancing a NearLink digital key solution, expected to be introduced in vehicles in 2030

October 28: Media learned from people familiar with the matter that BMW is advancing a NearLink digital key solution and intends to have STMicroelectronics and NXP supply multi-mode chips combining Bluetooth and NearLink for vehicle passive unlocking, with plans to introduce it in vehicles in 2030. (Lanjing Auto)

OpenAI plans to buy Google AI chips

October 28 news: According to foreign media citing people familiar with the matter, OpenAI plans to purchase Google's AI chips. The report did not disclose the transaction amount or further details. Google has not yet responded to a request for comment from foreign media, while OpenAI declined to comment.

If the plan is implemented, it will further deepen the cooperative relationship between OpenAI and Google's parent company Alphabet. The two companies are competitors in AI products and search business, yet they have successively reached commercial cooperation. Last week, OpenAI just announced that it would open its frontier large models to the outside world through Google Cloud.

As companies seek cheaper chips suitable for running artificial intelligence models, Google's tensor processing units (TPUs) are becoming increasingly popular as an alternative to Nvidia's market-leading AI chips. Google began selling TPUs directly to some customers this year. This move also reflects OpenAI's multi-pronged strategic intent to urgently improve its own AI computing power. (Sanyan Technology, Cailianshe)

SK Hynix says DRAM and HBM capacity is basically sold out for the next three years

October 28: SK Hynix said at its earnings conference that its DRAM and high-bandwidth memory (HBM) capacity is basically sold out for the next three years through 2029, and its long-term sales outlook has improved significantly. The company completed shipment of next-generation HBM4E samples in the second quarter, and HBM4 development and certification progress is in line with customer timelines. (Sina Finance)