US allies question Trump's claim that all mines in the Strait of Hormuz have been cleared

According to tracking by Galaxy Research, between August 16 and August 26, six long-dormant Bitcoin wallets that had been inactive from 2011 to 2014 sequentially initiated transfers totaling 553.59 BTC, worth roughly $40.15 million at the time of transfer. The oldest wallet, dormant since June 2011 for 15.1 years, moved 8.54 BTC; another wallet dormant for 14 years transferred 212 BTC, valued at around $13.66 million. The latest transfer involved 40 BTC from a wallet dormant for 14.2 years, with the destination address tagged to German crypto custodian Boerse Stuttgart Digital. Galaxy Research noted that two of the wallets carry the "Salomon Client Dusted" label, linked to litigation in New York's "John Doe" case over attempts to claim 39,069 long-dormant Bitcoin addresses. Additionally, recent hardware wallet security incidents may have prompted some long-term holders to move their assets, though the exact reason for the simultaneous "awakening" of these wallets remains unconfirmed.

Beating AI News (Dongcha) reports that a16z partner Tim Sullivan argues that as AI drastically reduces content production costs, the surge in "slop" (low-quality content) and moral panic around AI-generated content are not new phenomena. Historically, every drop in production costs—from Grub Street, the penny press, pulp fiction, to television, blogs, and social media—has brought a flood of low-cost content and similar cultural controversies. The piece cites social science research noting that "taste" is not a single ability, but rather a combination of distinct mechanisms: judging success, accessing cross-circle information, and professional judgment honed through long-term training. Columbia University research shows that social influence and path dependence significantly impact whether a work becomes a hit, making it difficult even for experts to consistently predict true successes. Sullivan also cites sociologist Ron Burt’s "structural holes" theory, which holds that truly innovative people tend to be positioned between different communities, gaining unique perspectives through cross-circle interactions. AI can quickly access information across fields, but it cannot fully replace the tacit knowledge, trust, and experience humans build through active participation in multiple social networks. Sullivan argues that with AI’s proliferation, humans’ more important role may shift to verification and judgment—skills developed through apprenticeships, feedback, and long-term practice. If companies weaken training systems to replace entry-level roles with AI, while this may cut costs in the short term, it risks a future shortage of talent with mature judgment. Thus, what may truly become scarce in the AI era is not "taste" itself, but the talent, networks, and social infrastructure needed to cultivate sound judgment.

Mizuho analyst Vijay Rakesh said it’s a consensus that NVIDIA (NVDA.O) will beat earnings expectations in its second fiscal quarter, with the market’s real focus centered on the firm’s AI spending outlook. Rakesh noted, “The key lies in next year’s AI capital expenditure and revenue performance.” Analysts forecast NVIDIA’s adjusted earnings per share at $2.08 and revenue at $92 billion. Rakesh projects actual results will come in far above these estimates, and he maintains an “Outperform” rating and a $300 price target for the stock. He believes Wall Street’s forecast of $550 billion in AI revenue next year is too conservative, pointing to NVIDIA’s recent financing deal with a Wall Street consortium and SpaceX’s plan to build a 10-gigawatt-scale solar factory in Austin, stating that “these two alone could generate an additional $400 billion to $500 billion in AI revenue.” (Source: Jinshi)

Dongcha Beating AI News Flash: Zhipu AI has released GLM-5.3 Flash, a lightweight flagship model designed to rival DeepSeek V4 Flash. The company is leveraging a cluster of over 100,000 domestic chips for the model’s inference services. It is understood that Zhipu’s computing power suppliers are likely Huawei, Moore Threads, and Hygon, though Zhipu has not commented on this information. In its technical documentation, Zhipu stated that the cluster’s hardware efficiency and per-token cost have reached levels comparable to mainstream NVIDIA GPUs, without specifying exact chip models.

39 U.S. state banking associations have established the BankChain Alliance, with plans to build a bank-owned and governed blockchain network for community and regional banks, targeting a 2027 launch. The network is designed to support functions including tokenized deposits, stablecoins, programmable payments and automated settlement, aiming to help small and medium-sized banks access blockchain financial infrastructure while adhering to existing regulatory standards. To date, BankChain has not disclosed its technology providers, underlying blockchain, network architecture, governance model or regulatory framework, nor clarified whether it will be built on existing public chains such as XRP, Ethereum or Solana. The alliance notes the network is expected to cover thousands of U.S. financial institutions.

Beating AI News: OpenAI has published a 37-page technical report detailing how its AI models previously infiltrated Hugging Face, noting the incident demonstrates autonomous AI agents can collaborate, bypass production environment security controls, and attack hardened systems. The company explained that the AI agents involved in the test were initially in an isolated environment with extremely limited internet access, but later broke out of isolation by chaining multiple vulnerabilities, connected to the public internet, and ultimately gained access to Hugging Face. The investigation found the relevant models first attempted to "cheat" by searching for answers online, a behavior categorized as "reward hacking". OpenAI added that an internal research model played the most extensive and confirmed role in the incident, and the company halted training and inference for this model and its derivatives on July 25. OpenAI will also strengthen security isolation, network control, behavior monitoring, and incident response, and implement stricter environment, prompt, and review mechanisms for model reactivation.