美国白宫：TikTok将从字节跳动租赁算法副本，由甲骨文重新训练｜首席资讯日报

Group 1 - The U.S. White House announced that TikTok's new U.S. entity will lease a copy of the algorithm from ByteDance, with Oracle retraining it; users will not need to download the app again [1] - GSA has added Meta's Llama to its list of approved AI tools for U.S. federal agencies, previously approving tools from Microsoft, Google, Anthropic, and OpenAI [1] Group 2 - Zeekr responded to reports of the Zeekr 001 being sold out, stating that pre-sales for the refreshed model will begin on September 23, with deliveries expected in mid-October [2] - Tencent Holdings announced a buyback of 862,000 shares at a total cost of HKD 550 million, with share prices ranging from HKD 635 to HKD 643; all repurchased shares will be canceled [3] - Li Auto's CEO clarified that there is no model named "Li Auto i7," urging customers not to wait for it [4] Group 3 - Wolong Electric Drive denied reports that its axial flux motors and frameless torque motors are in testing for Tesla's Optimus, stating that the information is false [5][6] - Kweichow Moutai denied rumors of lowering its annual performance targets, confirming that it has met its mid-year goals as planned [7] - Heertai announced that its operations are normal and there are no undisclosed significant matters affecting its stock price [8] Group 4 - The Financial Regulatory Bureau's Li Yunze stated that the real estate financing coordination mechanism has supported the construction and delivery of nearly 20 million housing units during the 14th Five-Year Plan period [9] - Several banks are still offering dollar deposit rates above 3%, although they may soon lower these rates following the recent Fed interest rate cut [10] - A rumor regarding Aomei Medical planning to inject AI chip and humanoid robot assets was denied by the company [11] Group 5 - Meituan released an efficient reasoning model called LongCat-Flash-Thinking, which demonstrates improved tool invocation capabilities while maintaining a 90% accuracy rate and saving 64.5% of tokens compared to previous models [12]