2026-11-07

Yangtze Memory Technologies plans a fourth fab in Wuhan, scheduled to begin construction in 2027, and is targeting domestic equipment at 60% of purchases, with a maximum of 70%, up from more than 50% at its third fab, Reuters reported. The plant would produce NAND flash with more than 300 layers for data-center and AI storage. These are plans rather than completed capacity: cleaning, polishing and etching tools are expected to supply much of the domestic increase, while lithography and inspection equipment would still depend on overseas suppliers. YMTC&\x27;s China NAND market share has risen to 38.5%, ahead of Samsung, and its total capacity could double once its new fabs are fully operational, according to the report.

The yield on CoreWeave&\x27;s 2032 bond, which carries a 9.625% coupon, has climbed above 16%, while its five-year credit default swaps—insurance against default—spiked above 1,289 basis points on Thursday, according to Deutsche Bank, as credit markets demand more protection from AI data-center borrowers.

The Nasdaq Composite fell more than 5% on Thursday and closed more than 11% below its record set last month — a decline greater than 10%, which Wall Street calls a correction. CNBC reported that across 10 prior corrections in the past decade, the index gained an average of 9.41% from its bottom over the following three months, while the 2022 episode instead brought a further 18.3% six-month loss.

Libghostty 1.0 has been released as the first stable version of Ghostty&\x27;s embeddable terminal-emulation library. The release makes Ghostty&\x27;s terminal emulation available for reuse in other applications, targeting developers who would otherwise need to build that functionality themselves. The supplied material does not detail supported platforms, APIs, or compatibility.

The PwnEscape Part 2 write-up walks through a practical escape from a Firecracker microVM by exploiting the already-disclosed PwnEscape vulnerability, presenting itself as an exploit-development guide rather than a new disclosure. The configuration tested had no Firecracker jailer; the write-up notes that production containment also drops privileges and restricts the VMM with namespaces and seccomp, so a jailed deployment needs a separate assessment of the privileges and resources available after VMM compromise.

Boson AI announced Higgs Realtime Duplex, a 45B mixture-of-experts multimodal model designed to process audio, video and text in 160 ms chunks while listening, watching and speaking concurrently. Unlike turn-based voice models, it keeps thinking, calling tools and delegating harder problems to a larger backend model as background spans in the same stream, and can begin speaking before staged tool results are complete. The model is not open weights: access to the Duplex API is waitlisted, while the earlier turn-based Higgs Realtime is available now. Boson&\x27;s reported comparison shows the model competitive with, but on five of seven benchmarks 1–5 points behind, OpenAI&\x27;s best settings, with turn-taking its weakest area.

Moonshot AI is reportedly negotiating to rent more than 8,000 Nvidia Blackwell GPUs from European neocloud data centers, mainly for inference and some training, according to CNBC&\x27;s account of unnamed sources. The arrangement would use remote access to work around U.S. export controls that bar Blackwell imports into China; the report says Moonshot may have to pay a 30%–50% higher prepayment than non-Chinese buyers. The White House said in July that the company had obtained Blackwell chips through a Thailand facility. The reported plan has not been confirmed by the companies involved.

Baidu has open-sourced Wenxin 5.5 Flash Base, a mixture-of-experts base model with 280 billion total parameters and 14 billion activated per token, according to Xinhua. The model was pretrained on 33.5 trillion tokens with a 128K-token context, and fine-tuning and inference are said to support contexts up to 1 million tokens. The report gives no license, repository link, benchmark results, or independent verification, so public availability and performance remain unconfirmed.

Taobao’s MNN team has reportedly open-sourced MNN-Pocket-0.3B, an approximately 300-million-parameter model designed to run fully offline on phones. The release includes separate versions for pretraining, post-training, and on-device deployment, with an INT4 deployment file said to be 173 MB and a target peak memory below 300 MB. The team reportedly used a 28-layer structure, cut the vocabulary from 151K to 64K, and claims average decoding of about 350 token/s on an iPhone 18 Pro, against a target of more than 200 token/s. Those speed and memory figures are source claims from a brief secondary report and have not been independently verified.

San Francisco Fed President Mary Daly said Friday she expects further interest-rate increases will likely be needed, citing persistent inflation; September consumer prices rose 3.7% from a year earlier, above the 3.1% annual rise in average hourly earnings in October.

rss · Sean Goedecke · Nov 7, 00:00 「Background」 Sean Goedecke reflects from frontend experience on why designers are unusually valuable yet culturally distant from engineers, and why scaling their work creates organizational friction. 「Solution」 His central practical claim is that a team should have at most one designer. He describes a product with three: after the original designer rotated off, two senior designers replaced them, redesigned the UI, then repeatedly reversed each other&\x27;s changes page by page until management froze the design near the deadline. He offers two speculative explanations: engineering disputes can lean on concrete measures such as latency or bugs, whereas design disputes rest on taste; and design changes are product-wide because users experience one interface, while many engineering choices remain invisible. Design systems, which he dislikes but considers better than the alternatives, suffer from misaligned incentives—owners want elegant reusable components while product teams need awkward customer requirements—and compatibility problems when the system targets a modern framework but production code mixes old technologies. Engineers then hack around it with CSS or reimplement components, and version differences create mismatches with designers&\x27; mockups. He also argues that some teams, such as APIs, billing, or rendering, lack a meaningful visual surface, and that poorly designed products can still succeed, so design is not always business-critical. 「Takeaway」 Goedecke&\x27;s conclusion is not that design matters less, but that its organizational cost is uneven: one embedded designer can transform a product, while multiple designers and rigid design systems can consume engineering effort without guaranteeing better outcomes. Tags: design systems, frontend engineering, team dynamics, design collaboration