OwlAuthorizationMonitor

Google DeepMind's reinforcement learning system can dynamically control data center cooling and workload scheduling, learning optimal strategies based on historical telemetry data, and operated under human supervision in a closed loop with safety constraints. The initial deployment achieved a double-digit reduction in cooling energy consumption while maintaining reliability and thermal limits. The system also introduces grid-aware scheduling, shifting flexible AI tasks to periods of low grid stress or high renewable energy availability. As a case study of applying RL in large-scale industrial systems, Google states it is willing to share experiences with operators. Privacy and environmental advocates are closely monitoring, and the industry is working to offset the energy consumption growth caused by AI workloads.

AI remains a long-term core opportunity, but short-term volatility is enormous. Concerns about AI’s rally turning into a bubble coexist in both global and domestic markets. Institutional selling off—driven by reductions in holdings—has caused some sectors to pull back, while retail investors continue to lose money. Despite differences in sentiment, Ma Huateng, Jack Ma, and others emphasize grasping AI’s long-term opportunity. The industry chain—from computing power infrastructure to large-model platforms to industry applications—still has broad room for growth, and the next round of opportunities will go to those who are well-prepared and keep a steady mindset.

According to SoSoValue, last week (5/25–5/29), Bitcoin spot ETF experienced net outflows of $1.42 billion, the third-highest in history. The largest weekly outflow was from IBIT, at $966 million, with a total net outflow of $63.81 billion; GBTC had a weekly outflow of $175 million, with a total net outflow of $26.62 billion. As of the time of writing, the total net asset value is $94.17 billion, the ETF net asset ratio is 6.38%, and the total net inflow in history is $55.66 billion.

Juan Manuel Sobral, Chairman of the Uruguay Blockchain Association, warned that the cryptocurrency regulatory project Uruguay is currently advancing has serious gaps. It not only fails to differentiate the risk levels of different business models, but also sets an admission guarantee of up to $100,000, which is extremely unfriendly to startups. Sobral said that the regulators “do not understand what they are regulating,” overlooking cutting-edge areas such as NFTs and RWA. He is concerned that this rigid regulation will force local innovative companies to seek operating space abroad, leading to losses for Uruguay at the national level in terms of tax revenue and industrial development. (CriptoNoticias)

Wintermute announces its entry into the prediction market sector, providing two-way liquidity and continuous quotes for event contracts on mainstream platforms. The related platforms have a monthly trading volume of over $20 billion. By April 2026, the cumulative historical trading volume of Polymarket and Kalshi exceeded $150 billion. Industry insiders believe that prediction markets have demand characteristics similar to large assets, but liquidity is still in the early stages. Ongoing bilateral liquidity helps narrow spreads, support large transactions, and improve price signal quality. Previously, Jump Trading and Galaxy Digital have also entered the space.

[ME AI]() Message, Google used its development tool Google AI Studio to create an online quiz about the major announcements at Google I/O 2026 through vibe coding. (Source: AiHot)

CryptoWorld News reports that the prediction market shows that Colombian presidential candidate "Tiger" De La Espreeira has promised to build 10 super prisons, sparking comparisons with El Salvador's CECOT. Currently, the market predicts a 67% chance of him winning the Colombian presidential election.

Mars Finance News, May 28 — "New Stock God" Serenity posted on social media saying, "Seeing so much support from Chinese communities on X really makes me happy. This reflects a very interesting cultural difference: people try to understand my thought process and stock selection logic, and use that to improve their own investment systems. In contrast, people from other cultural backgrounds might immediately dismiss everything. Maybe for fun, I will start writing opinions on two Chinese stocks, even if I don't hold any positions."

Bijie.com News reports that cellular semiconductor manufacturer Sequans Communications (NYSE: SQNS) has raised funds by selling nearly 80% of its Bitcoin holdings, completing the full remaining convertible bond redemption related to its Bitcoin treasury. At present, all of the company’s convertible bond redemption has been completed, and the remaining 658 Bitcoins have been fully released from any encumbrance. Sequans said it plans to gradually “liquidate” the remaining Bitcoins over time, but did not specify whether it will be sold directly, used as collateral, or used in other on-chain strategies.

Federal Reserve's Williams stated that how productivity improvements ultimately affect interest rates and monetary policy remains unclear, depending on the nature of the changes and the expected duration. Currently, officials are assessing the impact of recent productivity surges and higher productivity expectations brought about by artificial intelligence on inflation and the labor market, with many officials still uncertain about its future trajectory.

ME News Report, April 18 (UTC+8), according to CoinMarketCap market data, BTC fell below $76,000, currently quoted at $75,984, a 24-hour decline of 2.3%. (Source: CoinMarketCap)

Mars Finance news: The artificial intelligence coding startup Cognition announced the completion of a $1 billion funding round, with a valuation of $26 billion.

This article reviews MiniMax's M2 series technical reports, describing the trade-offs from M1's hybrid linear attention to full attention, as well as MTP, Sigmoid routing, and Forge in reducing costs on training and inference. It is the first disclosure of the Forge and M2.7 self-evolution mechanism for long sequence Agent RL, which uses windowed FIFO and prefix tree merging, achieving up to a 40-fold speedup in training on long sequences. The self-evolution closed loop of M2.7 can complete over 100 rounds of analysis, code modification, evaluation, and rollback, improving by approximately 30%. With a single token of 9.8B parameters, SWE-Pro achieves 56.22%, MLE Bench 66.6%, approaching Gemini 3.1.