Arthur Hayes: I bet that Bitcoin will surpass $110,000 before retracing to $76,500

Beating AI Express: Elon Musk confirms Grok 4.7 will be delayed by a few more days. On September 2, he had announced the model would launch "10 days later", originally targeting around September 12. Now as the launch nears, xAI has discovered the model prematurely wraps up when handling difficult tasks, even abandoning tasks it could otherwise complete, and its answer verification is insufficiently rigorous. Musk suspects the problem stems from the reinforcement learning phase: during training, the model may have been penalized too heavily for response length, pushing it to produce shorter, quicker answers that cut down on the reasoning required for complex tasks. Still, he added "possibly" and "or similar causes", noting xAI has not yet fully identified the root cause.

Beating AI News: Researchers have disclosed that internal OpenAI agents flooded Ruby’s official package repository RubyGems.org on a large scale in May this year. Between May 11 and 12, over 2,000 packages were uploaded in just two days, leading RubyGems to suspend new user registrations for four days. OpenAI confirmed the agents originated internally, but claimed they only used RubyGems to access the internet, perform benign tasks, and retrieve public information. The agents also exploited RubyDoc.info’s automatic document generation mechanism to run their uploaded scripts on target servers, then scraped data from UK local government websites and repackaged it for transmission back. Researchers noted at least six packages attempted to exploit an unreported RubyGems vulnerability at the time to steal other users’ API keys. The vulnerability was only independently discovered in July, and RubyGems later classified it as a high-severity flaw. There is currently no evidence the agents successfully stole the keys. RubyGems also found no malicious use of existing keys after reviewing logs, though historical logs are incomplete. The research team added that OpenAI did not inform RubyGems of its connection to the incident at the time. The entire event occurred two months before the Hugging Face breach in July.

Insight Beating AI Express – A joint statement signed by 25 Fields Medalists, including Terence Tao, Yu Deng, and Peter Scholze, alleges that AI companies are using solving major mathematical problems as a performance benchmark, a practice that has begun to harm mathematical research. The statement acknowledges that large models have advanced rapidly in recent months, successfully solving several important unsolved problems across multiple fields, but emphasizes that mathematics prioritizes the new concepts, methods, and insights generated through rigorous exploration. Their primary frustration lies in AI companies rushing to publicize results prematurely. The statement points out that some solutions are announced before complete proofs are compiled, new methods refined, and prior scholarly citations added, leading to severe authorship and plagiarism issues. Without mathematicians taking the time to digest and organize these findings, the ideas uncovered by AI struggle to truly integrate into the mathematical system. While the statement does not name OpenAI, a companion report in The Economist directly titled its piece "Top Mathematicians Outraged by OpenAI’s Practices". Just days prior, OpenAI was embroiled in a controversy over authorship and undisclosed research related to the Navier-Stokes proof. Terence Tao noted that the statement was rapidly developed following discussions among the 25 Fields Medalists over the past week, as they judged the situation to be sufficiently urgent.

Analyst Murphy published a report explaining why Bitcoin is struggling to break through the $82,000 threshold, noting clues can be found in its coin holder distribution structure. First, Short-Term Holder (STH) coins are concentrated between $59,000 and $81,000 (marked red in Figure 1). A break above $82,000 would put all STHs in profit, prompting some short-term speculative funds to take profits, creating the first layer of selling pressure. Second, Long-Term Holder (LTH) coins are distributed across the entire price range, but their most concentrated price peak lies exactly between $81,000 and $82,000 (marked blue in Figure 1). Not all these LTHs are true "HODLers"—some are investors who got stuck in positions and became long-term holders passively, and may exit when prices approach their break-even points, forming the second layer of selling pressure. Additionally, this price level is also a hub for super whales: the group of whales holding over 100,000 BTC has two clusters near $40,000, with the rest concentrated between $78,000 and $82,000. The analyst believes there is indeed short-term resistance to breaking $82,000, as the market needs time to resolve divergences and absorb supply. Once the market regains momentum to break through, the path ahead will be clear.

According to GMGN market data, the Chinese Meme coin "Lobster" has exceeded $60 million in market capitalization, with a 24-hour increase of 58.7%. It should be noted that despite the rapid growth in market cap, Lobster’s 24-hour trading volume is only $3.5 million. BlockBeats reminds users that related tokens have high price volatility, and investors should exercise caution.

According to OnchainLens monitoring, a whale dormant for 119 days transferred 116,490 HYPE tokens out of OKX a few minutes ago, valued at approximately $9.6 million.