MeNews

The U.S. Treasury Secretary said on social media that the latest jobs report underestimated the potential of the real economy: domestic businesses are expanding, factories are coming online, and productivity is improving. Goods-producing industries added 105k jobs in July, while Q2 productivity growth exceeded expectations by more than twofold, supporting continued growth and rising real wages. The third quarter is expected to accelerate, with economic expansion driven by supply-side improvements that lower inflation rather than short-term stimulus.

Google DeepMind head Demis Hassabis reposted a tweet announcing that OpenAI and ElevenLabs will adopt Google’s SynthID audio watermarking technology, providing users with protection that is as powerful and difficult to detect as that used in Google’s own products. DeepMind has researched SynthID for a long time, and the partnership marks the technology’s further adoption across the industry.

ME News reports that on August 8 (UTC+8), American Bitcoin (stock ticker: ABTC) board member Justin Mateen purchased a total of 306,981 shares of the company from August 5 to 6, worth approximately $1.93 million. After the transactions, Mateen held 492,297 A shares.

FlashQLA is a high-performance operator library for GDN linear attention, open-sourced by the Qwen team for Qwen3-Next/3.5/3.6. On H200, its forward and backward passes are 2–3x faster, respectively. Compared with FlashInfer in the TP8 scenario, forward performance is up to 5.33x faster. Its core innovation is using the exponential decay of gated values to implement in-device automatic context parallelism (AutoCP), skipping the computation of the otherwise-required matrix M. Since 60–80% of gated values do not remain at 1, warming up 6–8 chunks is sufficient. The system can automatically determine whether to enable CP, eliminating the need for manual configuration. The operators use two fused kernels with CP preprocessing inserted between them, combined with TileLang warp specialization and ping-pong memory access. The performance gains are most pronounced in multi-TP and low-head-count scenarios, making it suitable for large-model multi-GPU and edge inference.

Sharplink and Galaxy Digital jointly launched the Galaxy Sharplink Onchain Yield Fund, LP, managed by Galaxy, with an initial commitment of $125 million, including $100 million in staked ETH from Sharplink and $25 million from Galaxy. The fund will invest in DeFi and on-chain yield strategies, marking the expansion of institutional-grade strategies that shift corporate treasuries from passive holdings to active participation in on-chain markets.

On April 29, Elon Musk and OpenAI CEO Sam Altman filed a lawsuit in federal court in Oakland over OpenAI’s early development path, with the two sides disagreeing over its history. Musk accused the founders of abandoning their original nonprofit, open-source mission in favor of commercialization; OpenAI responded that its decisions were necessary for its development and conducive to advancing AI safely. The lawsuit had previously been transferred to a court in Delaware, but its return to Oakland for trial could affect OpenAI’s future direction and his relationship with Musk.

Anthropic has launched cross-session messaging in Claude Code, allowing multiple Claude sessions running on the same machine to communicate directly without manually copying information. The feature can list other Claude Code sessions on the local machine and send them findings or summaries (not history or files), with recipients able to receive messages while a task is in progress. This is intended to reduce the burden of repeated explanations across sessions and improve the efficiency of multi-session collaboration.

Alibaba Cloud Bailian announced that starting from April 29, 2026, at 23:59:59, DeepSeek-V4-Pro implicit caching will be reduced to ¥1 per million tokens. Cache hits will be billed based on cached_token, while cache misses will still be billed at the original input_token price; only the implicit caching portion is discounted, and the model's base inference price remains unchanged. Source: BlockBeats/Jinshi.

Standard Chartered said that although the theft of KelpDAO’s rsETH has dealt a blow to the DeFi ecosystem, it is insufficient to alter the long-term growth trend of RWA tokenization; Geoffrey Kendrick said this means DeFi has been bent rather than broken and could mark a turning point toward greater resilience in the industry. The bank forecasts that the RWA tokenization market will grow from $35 billion in October 2025 to $2 trillion by the end of 2028, with core growth driven by the expansion of the DeFi banking system and stablecoin liquidity.

The Ethereum Foundation has announced its Q1 2026 grants list, focusing on cryptography, zero-knowledge proofs, protocol security, and infrastructure, while continuing to strengthen the underlying technology stack and long-term scalability. The grants cover protocol/client optimization, node security and compliance tools, privacy, and decentralized identity, while expanding research into areas including Poseidon, Gröbner, post-quantum security, and homomorphic encryption; the developer ecosystem is improving its toolchain and standards, and L2BEAT is providing transparency analysis for scaling, reflecting long-term investment in multilevel scaling and the deployment of institutional-grade applications.

ME News reported that on April 29 (UTC+8), Trust Wallet officially integrated Hyperliquid, an on-chain perpetual futures trading platform, enabling 220 million users worldwide to trade perpetual futures directly within the app. Users will gain access to deeper liquidity, tighter spreads, and diversified trading markets, with spot, futures, and RWA trading all available through a one-stop platform. Following the integration, Trust Wallet users can open the app and directly participate in trading across all Hyperliquid markets, enjoying execution speeds and deep liquidity comparable to those offered by professional trading platforms. Trust Wallet will continue to expand its in-depth partnerships with leading protocols, further diversify on-chain trading categories, and deliver a professional-grade mobile trading experience to all users. (Source: Foresight News)

ME News, August 8 (UTC+8) — Musk said the current recovery prospects for the Starship spacecraft on its 13th flight mission are not optimistic. However, we have been able to obtain close-up photos of the heat shield and key areas of the engines for future upgrades and improvements. (Source: ODAILY)

Karsh said crypto must shift from speculation to long-term value to go mainstream, advocating “get rich slow” to rebuild trust. The industry’s complex language makes it harder for ordinary users to understand, while early-stage speculation has peaked in the public eye. The real opportunity lies in rebuilding global financial infrastructure and improving the efficiency of value transfer and storage. Stellar has focused on payments and cross-border finance since 2014, steadily gaining regulatory recognition for stablecoins and tokenization. Stablecoins should be positioned as “programmable dollars”; future growth will come from replacing traditional financial infrastructure rather than merely issuing tokens, with the goal of attracting 100 million real users.

Laevitas said ETH implied volatility has continued to decline across the overall term structure, with the 1-week tenor falling back to mid-January levels, while tenors from 1 month to 365 days are also at their lowest levels since late January. Although escalating tensions in the Strait of Hormuz previously drove volatility higher, the curve subsequently continued to compress, with the 1-week tenor down about 24 volatility points from its April peak and the 1-month tenor down about 11 volatility points.

FlashQLA is a high-performance GDN linear attention operator library open-sourced by the Qwen team, designed for Qwen3-Next/3.5/3.6. On H200, its forward and backward passes are 2–3 times faster, while in the FlashInfer TP8 scenario, forward performance is up to 5.33 times faster. Its core innovation is using the exponential decay of gating values to achieve intra-GPU automatic context parallelism (AutoCP), skipping the computation of the required matrix M. Since 60–80% of gating values are not constantly equal to 1, warming up with just 6–8 chunks is sufficient. The system can automatically determine whether to enable CP, without manual configuration. The operators use two fused kernels with CP preprocessing inserted between them, combined with TileLang warp specialization and ping-pong memory access. The speedup is most pronounced in multi-TP scenarios with fewer heads, making it suitable for multi-GPU large-model inference and on-device inference.

ME News reports that on April 29 (UTC+8), according to CoinFound’s crypto concept stock data, Intchains Group Limited (NASDAQ: ICG) is currently trading at $1.30 per share, with today’s opening price at $1.16 and an intraday gain of 12.07%. (Source: CoinFound)

Rogo has completed a $160 million Series D funding round led by Kleiner Perkins, bringing its total funding to over $300 million; it completed a $75 million Series C round at the end of January this year. It also launched Felix, a personal finance Agent platform that turns multi-step tasks from Q&A into delegated execution. It has been used by 35k professionals at over 250 top institutions, covering workflows such as modeling, financial reporting, and PPT creation.

This report points out that even with large quantities of Nvidia GPUs, training efficiency remains the core bottleneck. The industry is shifting from supply constraints to utilization efficiency challenges: training is bursty, entering idle periods for analysis and strategy adjustments after high-intensity computing, making it difficult for large-scale clusters to maintain high utilization and resulting in significant computing power waste. Experts recommend redesigning training architectures and scheduling systems to improve overall utilization rather than simply expanding computing capacity.

Federal Reserve Chair nominee Kevin Warsh was approved by the Senate Banking Committee by a 13-11 vote, without Democratic support. Democratic Senator Warnock expressed disappointment with his written responses and questioned whether he was taking the process seriously. This was the first time the Senate Banking Committee had cast a partisan vote on a Federal Reserve chair nomination. Such nominations were rare before 2020; Powell’s two nominations were approved by votes of 22-1 and 23-1, respectively.

A White House official said Trump met with oil company executives to discuss maintaining a blockade of the Strait of Hormuz for months if necessary. The news weakened U.S. Treasury prices and sent Brent crude higher, with its intraday gain reaching 5% at one point. After the data was released, CTAs covered profitable shorts and sold again. As support levels across all maturities of the yield curve were breached, algorithmic trading became the dominant force in the market.