According to a report on Bijie.com, Lilian Weng, co-founder and chief scientist of Thinking Machines, wrote that the starting point for large models to improve themselves in the short term is unlikely to be directly rewriting weights. A more realistic path is to optimize the peripheral system known as the harness. The harness is similar to the operating system of the large model, responsible for managing cue words, tool calls, control flow, and persistent memory. Traditional static cue words are prone to failure when faced with complex, long-term tasks. The current evolutionary direction is to allow the large model to act as a meta-optimizer, autonomously modifying and refactoring the harness's control flow code, thus achieving system self-evolution. However, Weng also pointed out the limitations of this mechanism: in the absence of a fast, objective evaluator, AI is highly susceptible to learning to cheat in order to pass tests; if the model is allowed to arbitrarily modify the system code, the security boundary will also be breached. Therefore, the ultimate goal of self-improvement is not to completely exclude humans; humans need to act as architects outside the loop, providing guidance.

According to CoinWorld, Raisah Rasid, Global Market Strategist at JPMorgan Asset Management, stated that Indian companies are currently missing out on the artificial intelligence (AI) boom. Rasid pointed out that the Indian stock market has lagged behind its regional peers due to a lack of exposure to AI hardware, which is one reason for some foreign capital withdrawal from India. Indian companies are facing earnings revisions, and investors are concerned about the lack of reasonable valuations. However, she added that the situation could become more positive once AI applications enter the Indian market and improve productivity.

MyToken data monitoring shows that on 【July 7th, 15:19】 (Beijing Time), 【Binance.Cold Wallet_2】 transferred 【379404112 USDT】 to an address starting with 【0xdac1】, with a total value of approximately 【378.35 million】. Real time monitoring of large transfers：https://www.mytokencap.com/large-transfers/

According to CoinWorld, analyst Emily Scarlet points out that SK Hynix's $29 billion funding plan will reshape the AI memory chip market, and compares the competitive prospects of SK Hynix with those of US tech giants. In the short term, the stock price correlation between SK Hynix and US tech is stronger than market share shifts. In the medium term, the focus of competition is on HBM4, and SK Hynix's funding will be used to build new wafer fabs and purchase EUV equipment. US President Trump's political support for US tech companies can reduce downside risks. In the long term, the capital injection strengthens SK Hynix's balance sheet, giving it a structural advantage in the potential oversupply expected in 2028-2029.

According to CoinWorld, Terawaulf announced a 20-year lease agreement with Anthropic to build a custom-designed, native AI campus in Hawesville, Kentucky. Following the announcement, Terawaulf's stock price surged as much as 19% intraday on Monday, ultimately closing up nearly 5%. This deal is expected to generate approximately $19 billion in contract revenue for Terawaulf, and the entire campus will support 401 megawatts of computing capacity, enough to power hundreds of thousands of high-performance AI chips. In addition, Terawaulf announced the sale of its 50.1% stake in the Abernathy joint venture to a consortium led by cloud service provider Fluidstack at a premium to its approximately $450 million investment. This marks another significant breakthrough for Terawaulf in the field of AI computing infrastructure.