It is reported that OpenAI Astra has been delivered to partners for testing, and will expand access further as early as next week.

Beating AI News: Liu Xing, head of public relations at Doubao, publicly questioned the claim that "over half of China’s AI computing power is used to produce short dramas", bluntly calling the relevant data "somewhat unreliable". He revealed that only around 10% of ByteDance’s total computing power is allocated to the training and inference of video generation models; this proportion also covers model R&D, not all of which is used for making short dramas. The controversy originated from an investor exchange record disclosed by Chinese online content provider China Digital Online in June this year. The company stated that 55% of domestic AI Token consumption comes from short dramas and video generation, 24% is used for e-commerce marketing, and software development accounts for only 15%. This data was recently resurfaced, and after media dissemination, it evolved into the narrative that "half of China’s computing power is used to produce short dramas", even sparking discussions such as "America builds the future, China builds short dramas". However, China Digital Online did not disclose the original statistical body, sample scope, and calculation method for this nationwide data. More importantly, AI Token consumption is not equivalent to actual computing power consumption, and not all video generation is for short dramas.

Affected by falling share prices of firms including Samsung Electronics and SK Hynix, South Korea’s primary pension fund, the National Pension Service (NPS), has suffered losses of 182 trillion won, equivalent to roughly 137 billion U.S. dollars. According to financial data provider FnGuide, as of October 15, the valuation of NPS’s portfolio holding stakes of 5% or more in 287 listed companies has dropped from 515.93 trillion won at the end of June to 333.53 trillion won. This marks the first quarterly decline for the fund since the fourth quarter of 2024; for NPS, which notched a record high return in the first half of the year, the slump represents a sharp reversal.

Ethereum co-founder Vitalik Buterin has published a post noting that placing most cryptographic support in precompiled contracts and the Ethereum Virtual Machine (EVM) is not an ideal solution. As data handled by both is typically written to Ethereum blocks via calldata, on-chain data space is limited. Even with the introduction of PeerDAS and Blob mechanisms, actual data throughput may still face constraints once it hits around 8 MB/s. Ethereum should not be viewed merely as a blockchain, but as a "cryptographic world computer". For computational tasks that do not need to be included in the main on-chain data and execution flow, they should not be forced to be processed in the main flow. A more appropriate approach is an EIP-8288-centric design: perform complex computations on the client, generate STARK proofs, then submit the proofs via EIP-8288’s recursive aggregation mechanism. Under this model, only the completion of computations and their results need to be verified on-chain, eliminating the need to directly process relevant raw data or execute the full computation process. Vitalik argues this will enable developers to freely use diverse cryptographic technologies, while keeping the Ethereum chain lightweight and reducing the requirement for adding new precompiled contracts in the future.

According to Bitget's market data, South Korea's KOSPI index posted an intraday gain of 1.00%.

According to monitoring by Onchain Lens, a crypto whale has staked a total of roughly 500,000 HYPE tokens over the past 24 hours, valued at approximately $38.92 million. The latest stake alone amounts to 399,994 HYPE tokens, worth around $30.92 million. Notably, this whale had previously unstaked its holdings two months ago.

Beating AI News Flash: Before Character.AI filed for bankruptcy, Disney had already extended job offers to 66 of its employees. 56 have left the company, while 10 are awaiting work visa adjustments. According to bankruptcy court documents, Disney plans to invest approximately $180 million in these employees over the next three years, including $30 million in retention bonuses. In comparison, Disney’s bid for some of Character.AI’s technology and patents is only $20 million, in exchange for permanent, non-exclusive usage rights. The agreement explicitly excludes existing apps, brands, user chat logs, and finished models; Disney does not intend to take over the entire chat platform. When Character.AI filed for bankruptcy, it only had 18 full-time employees left, 10 of whom are still waiting to join Disney. The company expects only 8 employees will remain through critical bankruptcy stages, with around 6 handling security-related work. Due to staffing shortages, Disney will also temporarily recall some employees who have transitioned roles to help maintain operations on the original platform. Meanwhile, 40 U.S. state attorneys general have intervened in the bankruptcy process, demanding protection of users’ private chat logs to prevent the data from being misused after asset sales. Other bidders may submit acquisition offers by November 20. If no party agrees to continue operations, Character.AI will shut down its services.