Arthur Hayes’ Flop Labs plans to build a computing power trading market enabling AI agents to directly purchase GPU computing power.

Germany’s 10-year government bond yield rose to its highest level since April 2011, climbing 3.9 basis points on the day to 3.377%.

Japanese financial brokerage firm Tokyo Tanshi said the probability of the Bank of Japan (BOJ) raising interest rates in September has risen to 97%.

Beating AI News – Tencent’s WorkBuddy Open Platform has officially launched, rolling out its AI agent capabilities to hardware manufacturers, industry application providers, and developers. The first phase offers five core capabilities: Buddy apps, expert tools, Skills, connectors, and hardware. Over 100 co-creation partners have already joined the platform. Hardware integration is the most prominent update this time. The first batch includes 9 co-branded smart hardware products, with partners such as Plaud, Rokid, Insta360, iFLYTEK, Anker, MOMA, and JD Jingzao. Taking Rokid’s AI glasses as an example, demonstrated use cases include data organization, market trend tracking, information lookup, and weekly report drafting. The glasses not only handle Q&A tasks but also act as a task entry point for WorkBuddy. Additionally, over 30 industry applications have been integrated, covering more than 20 sectors including finance, law, healthcare, education, and public welfare. Tencent launched the Buddy AI Ecosystem Co-Creation Program back in June, which initially had only 23 partners. Now with the official platform launch, the partner base has expanded to over 100. Liu Yi, Vice President of Tencent Cloud and Head of CodeBuddy and WorkBuddy, stated the goal is to build WorkBuddy into an "operating system for the agent era".

Kalshi, Polymarket, and Polymarket U.S. combined recorded total trading volume of $45.33 billion in August, a 14.5% month-over-month decline, marking the first monthly drop since August 2025. Kalshi’s August volume stood at $37.17 billion, down 7.3% from July’s $40.1 billion; Polymarket and its U.S. platform together notched $8.16 billion in August, a 36.7% decrease from July’s $12.89 billion. Despite the overall pullback, August’s trading volume remained notably higher than May’s $25.66 billion. Earlier forecasts had predicted that prediction market trading volumes would get a lift from the World Cup this summer, with the tournament running from June 11 to July 19, which spurred a clear rise in market activity. Meanwhile, Kalshi and Polymarket are now facing additional scrutiny from U.S. state regulators; more than a dozen states have taken enforcement actions or filed lawsuits against the two platforms, with disputes focusing mainly on sports-related contracts.

According to Bitget market data, the Nikkei 225 index closed down 1,889.70 points on Wednesday, September 2, a 2.85% drop, at 64,325.64 points. South Korea’s KOSPI index closed down 273.08 points on Wednesday, September 2, a 3.99% decline, at 6,562.72 points; SK Hynix fell 4.7%, while Samsung Electronics dropped approximately 4%.

Beating AI Express Insight: OpenAI Chief Scientist Jakub Pachocki has refuted some external interpretations of the Astra recurrent architecture. He stated that for OpenAI’s current frontier models, including Astra, the depth of their computational graphs is at most only twice that of GPT-4, rather than having suddenly increased by several orders of magnitude. This is a direct response to a previous report from The Information, which revealed that Astra uses recurrent depth, allowing the same set of Transformer layers to compute repeatedly. As a result, part of the inference process can occur internally within the model, without needing to be fully written out as Chain-of-Thought (CoT), sparking concerns about whether models will become increasingly difficult to monitor. Pachocki acknowledged that CoT monitoring is indeed fragile and is moving toward a direction that is harder to oversee. However, he clarified that this is not due to changes in model architecture. OpenAI has retained and used CoT monitoring since its first-generation reasoning models, and currently still lists enhancing this capability as a core research goal.