ARAMCO CEO said protecting the energy infrastructure and shipping routes the world depends on requires collective international cooperation.

人工智能研究員內森·蘭伯特在其Interconnects通訊中寫道：AI訓練和推理棧的很多環節都是高度可驗證的。訓練指標比如每GPU每秒token數，本質就是訓練速度。推理指標則是每prompt的token數、每token的FLOPs，說到底就是每個答案的成本。這兩塊都非常可優化，尤其是有成熟的子問題和架構權衡可以利用。我預計幾年內AIagent會幫助端到端優化整個流程，讓我們的推理能力逼近GPU這類加速器的底層算力上限。過去幾年，各家在推理效率上的收益已經很大了——比如模型以某個定價發佈後，還能再省下10–30%的服務成本。整個棧會持續複利，我預計未來幾年模型智能的有效成本會近乎指數級下降（可能比近期趨勢更快）。 這些效率方面的低垂果實，幾年內就能摘完。更長期的是加速器與模型的協同設計，時間線更長，能在靈活的GPU平臺之上再帶來幾個數量級的效率提升。在效率快速提升的這段時間裏，GPU的靈活性會是好事，因爲架構層面的探索空間很重要，是自動化研究的一個主要槓桿。在我看來，預訓練研究（至少是服務於當前這類模型的架構和數據選擇）2–3年內實現自動化，這個預測是合理的。

Oct 12 — At a regular Foreign Ministry briefing, CHINA responded to a Refinitiv question about a U.S. State Department statement last week that Chinese maritime activity off Taiwan's east coast "seriously undermines stability." CHINA said it has repeatedly stated China has an exclusive economic zone and continental shelf east of Taiwan and that Chinese activity there falls within its sovereign rights and jurisdiction, is lawful, and not open to criticism. CHINA said it is not the party provoking incidents or undermining stability in the area and reiterated that the Taiwan issue is an internal matter; it urged the United States to adhere to the one-China principle and the three China-U.S. joint communiqués, handle Taiwan cautiously, and take concrete actions to safeguard China-U.S. relations and regional peace and stability.