《晚安咖啡》科技夜话（英文版）：奥特曼宣布4个月后交付AGI，长鑫存储起诉五角大楼
晚安咖啡GoodNightCoffee
Hi there, good evening! Today is Sunday, August 30, 2026. I'm An An, your host from Good Night Coffee. After a busy week, tonight I'd love to chat with you about the latest in business and tech, and catch those interesting changes together. So relax, and let me help you wrap up your day.
@@@
According to 36Kr on August 30, OpenAI CEO Sam Altman announced that AGI will be delivered in four months, calling this goal the "final battle." Based on this timeline, AGI is expected to be completed by the end of this year. Altman set a specific deadline in his public remarks but did not reveal further technical details, including the system's capabilities, product form, or follow-up release plans. For now, information about the delivery remains limited, and further updates are expected to be shared later.
@@@
According to Readhub, ChangXin Memory Technologies has filed a lawsuit against the U.S. Department of Defense over its inclusion on the so-called "Chinese military companies" list. Headquartered in Hefei, Anhui, ChangXin is a semiconductor company focused on the R&D and production of DRAM chips. In December 2024, the U.S. Department of Defense placed ChangXin on the 1260H list under relevant legislation — the so-called Chinese military companies list. The company subsequently disputed the designation and formally filed a lawsuit in court at the end of August. Under current U.S. rules, companies on the list are subject to restrictions related to U.S. Department of Defense contracts and other measures.
@@@
According to Readhub, Apple recently released a new study on the relationship between large language models' reasoning abilities and training language. The results show that training for reasoning in Chinese alone achieves nearly the same results as training in English. This conclusion is based on systematic comparisons by Apple's research team of model reasoning performance across multiple language settings. The tests covered various reasoning tasks, including mathematical logic and common-sense judgment. The study found that when models complete reasoning training on Chinese-language data, their ability transfers stably to English and other language tasks, with no obvious language dependence or performance drop. The research team noted that reasoning abilities may be represented more abstractly inside the model rather than being tightly bound to specific surface language forms. The study focuses on the real practical impact of language choice on chain-of-thought training
effectiveness. The test data and method details have been published in the relevant technical report. These results offer useful insights for developing multilingual AI models, suggesting that when training resources are limited, building model reasoning capabilities does not have to rely heavily on English data. Apple says it will continue exploring how different language training strategies affect overall model ability.
@@@
According to Readhub, OpenAI has announced it will stop providing computing power support to the AI coding tool Cursor, a decision that directly affects Cursor's model-calling services. Shortly after OpenAI's decision, Anthropic said it would continue increasing computing power support for Cursor. Anthropic emphasized that it will ramp up resource investment to ensure Cursor's platform stability and service capacity remain unaffected. This means that after OpenAI exits the supply chain, Cursor can still rely on Anthropic for computing power. Cursor is an AI-assisted programming tool based on large model capabilities, which previously connected to services from multiple model providers. OpenAI's move mainly affects the model-calling layer, while Anthropic's additional support will to some extent fill the gap. So far, OpenAI has not provided further explanation regarding follow-up arrangements after the cutoff, and Anthropic has not disclosed the specific scale or timeline of the new
computing power support. The dynamics of cooperation and competition between the two companies in AI coding services are now shifting as a result of this event.
@@@
According to Bloomberg, some Micron Technology employees are threatening to strike, demanding that the company share the rich rewards from the AI boom. These employees point out that as demand for AI chips surges, Micron's high-performance memory chip business has seen both higher volumes and prices, yet ordinary employees have not received corresponding compensation. Employees involved in the action say they are not against the company's business growth, but want a fairer share of benefits during the AI industry's explosive phase, including higher pay and improved equity incentive plans. Micron, as a major global memory chip supplier, has its products widely used in AI servers and data centers, and the company has reported strong recent financial performance. At present, the employees involved are working on specific negotiation strategies, and whether the strike threat materializes still depends on follow-up communication between the company and employee representatives. Micron has
not yet made an official response.
@@@
According to foreign media reports, the South Korean government plans to offer generative AI services to all citizens for free, with no cap on token usage. This move means the country will fully popularize access to generative AI technology, and ordinary members of the public will no longer face usage limits when using related services. The open policy being advanced by South Korea covers all citizens, with the goal of allowing a wider population to access and use cutting-edge AI technology without barriers. Notably, the policy explicitly removes restrictions on token quantity, which differs significantly from the metered or tiered quota models commonly used by most international AI service providers. Analysts point out that by removing usage barriers and concerns about data volume, South Korea aims to drive deeper adoption of AI technology across all levels of society. Further details — such as which institutions will provide the service, what technical foundation will be used, and
the exact launch date — have not yet been announced.
@@@
According to Sina Tech, China has achieved a major breakthrough in bidirectional high-speed laser communication between the Earth and the Moon, successfully transmitting an 8K ultra-high-definition lunar image in just 12 seconds. The transmission distance was approximately 350,000 kilometers, and the communication rate reached 10 gigabits per second, marking a key step forward in China's deep-space communication capabilities. In the experiment, a ground station in Changbai Mountain, Jilin Province, transmitted laser beams to a spacecraft in lunar orbit, establishing a bidirectional high-speed laser communication link and sending the 8K lunar image back in real time. Laser communication offers significant advantages such as wide bandwidth, high speed, and strong anti-interference capability. Compared with traditional microwave communication, data transmission efficiency is several times higher, making it especially suitable for future crewed lunar missions and deep-space exploration
requiring high-bit-rate data transmission. This success verified the feasibility and stability of an ultra-long-distance laser communication link between Earth and the Moon, providing solid technical support for high-definition video transmission and scientific data return in future deep-space missions. The research team overcame several technical challenges, including atmospheric turbulence interference on laser signals and ultra-long-distance beam alignment, achieving high-precision acquisition, tracking, and pointing. This breakthrough also lays an important communication foundation for China's planned lunar research station and crewed lunar exploration program.
@@@
According to Reuters, Sony Music and Warner Chappell Music have jointly filed a lawsuit against artificial intelligence company Anthropic. The plaintiffs accuse Anthropic of using copyrighted lyrics on a large scale without authorization to train its AI model Claude, and claim that when users make related requests, the system can generate lyrics highly similar to existing songs. Sony Music and Warner Chappell believe this conduct constitutes a direct infringement of music copyrights and harms the legitimate rights and interests of songwriters and copyright holders. The lawsuit documents state that Anthropic did not obtain necessary licenses or pay the corresponding licensing fees during model training. The plaintiffs also allege that the AI tool has copied or closely paraphrased protected lyrics in its generated text. Based on these facts, Sony Music and Warner Chappell are asking the court to order Anthropic to stop the infringing activities, provide details of its use of training
data, and pay appropriate financial compensation. Anthropic has not yet formally responded to the lawsuit, but the company previously stated that its models include safeguards designed to avoid outputting copyright-protected material. The case is now drawing significant attention from both the AI industry and the music industry, and the court will conduct further hearings.
@@@
Alright, that's all for today's episode! Thank you for staying with me tonight. I hope this warmth carries you into sweet dreams. If you enjoy our show, you're very welcome to subscribe and share it with your friends — it really helps us a lot. See you next time!
