読書リサーチレーダー投資フレームワーク
ログイン / 新規登録
すべて技術製品投資株式市場その他

読書

2026-10-29198 投稿

更新

Li Auto Launches 2026 i6: Self-Developed Mach Chip and Lianshan Battery Enter the 250,000 RMB Market Segment

トピック · 理想马赫芯片融资重大
2026-10-29 08:59 GMT+8

On October 29, Li Auto launched the 2026 model year i6, with the most significant change being the integration of its self-developed Mach M100 autonomous driving chip and Lianshan 5C ultra-fast charging battery into this flagship pure electric vehicle. The company announced that the standard configuration price remains at 249,800 RMB, identical to the 2025 model, with a 10,000 RMB cash discount available during the initial sales period.

The Mach M100 is an automotive-grade chip independently developed by Li Auto, utilizing a 5nm process with a nominal computing power of 1,280 TOPS per unit. The chip was first deployed in higher-end models such as the L9 Livis this year; its entry into the i6 marks Li Auto's move to control the architectural design of its assisted-driving hardware and achieve deep software-hardware co-optimization in the mainstream 250,000 RMB market segment. Meanwhile, the new car features a 96 kWh self-developed Lianshan battery covering key components such as cells, packs, and BMS, increasing the CLTC range of the rear-drive version to 780 kilometers.

For Li Auto, which is accelerating its transition to pure electric vehicles, the i6 carries significant commercial weight. Financial reports show that in the second quarter of 2026, Li Auto recorded a net loss of 1.705 billion RMB, with its automotive business gross margin dropping to 9.4%, a decrease of 10 percentage points from the same period last year. By applying self-developed technologies to the i6, which has accumulated nearly 190,000 deliveries, Li Auto aims to dilute fixed R&D investments through sustained sales volume and improve platform reuse efficiency.

Additionally, Li Auto unveiled its European strategy, planning to enter the European market in the first quarter of 2027, with the i6 (named Li 6 overseas) launching first in Norway and the Netherlands. Li Xiang, founder and CEO of Li Auto, stated that the goal is for the i6 to become the company's best-selling model globally.

ソース:量子位

リサーチ

ProseMirror Product Developer: Under Tree-Based Document Architecture, Your Markdown Source Is Lost on the First Save

トピック · ProseMirror重大
2026-10-29 11:29 GMT+8

A developer with three years of experience working on ProseMirror-related products published a long-form post on V2EX, reviewing the source code of mainstream Markdown conversion libraries version by version. They present an architectural judgment: when the authority of a document lies in its JSON tree and Markdown is merely one way to print it, the source code you wrote is already gone by the time you hit save for the first time.

They cite examples from the most commonly used implementations. prosemirror-markdown 1.13.4 hardcodes italics and bold as asterisks; underline emphasis you write changes after one save. Unordered lists default to asterisks, so your habit of using hyphens disappears. Ordered lists are renumbered, and line-ending double spaces for line breaks are uniformly changed to backslashes. If it encounters syntax it doesn't recognize, it throws an error; the default schema lacks tables, footnotes, and strikethrough. Tiptap's official @tiptap/markdown 3.30 similarly hardcodes bold and italic formatting. Nodes without corresponding renderers output empty strings, causing content to silently disappear. The community-maintained tiptap-markdown has ceased maintenance; when encountering nodes it cannot express, it outputs placeholders. Tables with merged cells degrade to HTML, losing column alignment information.

The author emphasizes these are not bugs but inevitable results of the architecture. Since there is only one printing method, the source code you wrote is already gone upon the first save. This might not matter for personal note-taking, but it is critical for three scenarios: documents stored in Git show full-screen format changes even when modifying a single character; AI modifies only one section but normalizes the writing style of the entire document; the same file needs to be used across VS Code, static site generators, and other Markdown tools.

The author then tests the actual limitations of Markdown syntax itself. Using micromark 4.0.2 with GFM extensions, they tested item by item: table cells can only contain inline content; placing a list or code block inside will break the table immediately. Bold text cannot be nested within inline code, nor can links be nested within links. Chinese punctuation causes bolding to fail—CommonMark determines whether asterisks can start or end emphasis by checking if the surrounding characters are punctuation or whitespace. Chinese quotes and parentheses are punctuation, followed immediately by Han characters, so the rule deems them invalid. English writers rarely encounter this issue, but Chinese writers face it daily.

The author's solution is MSFE (Markdown Source-First Editor), making the Markdown source itself authoritative. Complex blocks store data within code fences, rendering as editable multi-dimensional tables within MSFE, while appearing as plain code segments in GitHub or VS Code that do not break or get lost. Collaborative editing operates on the Markdown source itself—a single Yjs text object. Since any segment of text is valid Markdown, there is no issue of deleting nodes due to structural invalidity.

ソース:V2EX

リサーチ

Harvard Professor Jeffrey Frankel: Purchasing Power Parity Is Unsuitable for Measuring Economic Power, ICP Data Revisions Far Exceed Exchange Rate Fluctuations

重大
2026-10-29 08:28 GMT+8

When comparing the economic scales of the United States and China, the choice between using purchasing power parity (PPP) or market exchange rates directly determines which country is ranked number one in the world. Under the PPP metric, China surpassed the United States in 2017; under the exchange rate metric, the US economy remains 47% larger than China's. Harvard Kennedy School professor Jeffrey Frankel wrote an article on EconBrowser explicitly opposing economist Paul Krugman's advocacy for prioritizing PPP, arguing that exchange rates are a more reliable tool for measuring geopolitical economic power.

Frankel's first line of argument targets the essence of economic power. He contends that a nation's economic power is manifested in its ability to impose trade sanctions, financial sanctions, gain leverage in international negotiations, or build military bases in Pacific island nations. These actions all take place in international markets, where the key factor is how much foreign assets a currency can purchase at the exchange rate, not how many goods it can buy domestically. China's capacity to mobilize resources on the global stage depends on the market exchange rate of the renminbi against the US dollar, rather than its domestic purchasing power adjusted for PPP.

Frankel's second line of argument is more technical. Krugman's primary defense of PPP is that exchange rate volatility is too high to make suitable cross-year comparisons. Frankel acknowledges this point but immediately follows up by noting that the revision magnitude of PPP data from the World Bank's International Comparison Program (ICP) actually far exceeds the annual volatility of exchange rates. He cites specific figures: the 2005 baseline data was revised downward by 40% during the 2011 survey; the extrapolated values based on the 2011 baseline were revised upward by 21% in subsequent surveys; the 2017 data was revised downward by 16%; and the 2021 data was revised upward by 5%. These revision magnitudes are substantially larger than the annual fluctuations in nominal exchange rates.

More unsettlingly, the baseline data itself is subject to modification. China's 2005 PPP exchange rate was initially published as 3.45 yuan per US dollar, but was revised to 2.86 yuan upon the release of the 2011 ICP data, raising China's 2005 PPP GDP by 20.5%; according to current data, this figure remains 21.8% higher than the original estimate. Thailand experienced a similar pattern: a 20% downward revision in 2005, a 43% upward revision in 2011, a 3% downward revision in 2017, and a 7% upward revision in 2021. In contrast, ICP revisions for advanced economies are much smaller because EU member states conduct new surveys every three years through Eurostat and the OECD.

Frankel concludes that if the sole reason for choosing PPP is exchange rate volatility, then the same criticism applies equally—and even more severely—to PPP data. For comparisons of national economic scale—especially those involving geopolitical power—exchange rates remain the more reliable choice. This implies that the US economy is still significantly larger than China's, and that US productivity growth continues to lead Europe's. Frankel also admits that PPP is the correct tool for comparing per capita income and living standards; his opposition is limited strictly to the specific purpose of measuring economic power.

ソース:econbrowser.com

リサーチ

Anthropic Letter to CA Governor Proposes Court Power to Block High-Risk AI Models

トピック · 加州AI紧急停止开关监管重大
2026-10-29 12:13 GMT+8

According to a report by INSIDE, artificial intelligence company Anthropic submitted a letter of opinion to California Governor Gavin Newsom on October 15, 2026. The letter formally responds to Executive Order N-9-26, signed by Newsom in September, which aims to establish regulatory frameworks for AI "kill switches" and safety oversight.

In the letter, Anthropic argues that when reviews determine an AI model poses severe risks, the government should be able to request courts to block its deployment. Specifically, Anthropic proposes legislation requiring large advanced AI developers to publish risk reports at least every six months, reviewed by Independent Verification Organizations (IVOs). If an IVO identifies risks and notifies the State Attorney General, and the AG determines the model poses a "risk of imminent catastrophic harm," they can petition a court for an order to suspend the model's deployment, internal use, or further development.

Anthropic stated that requiring both IVO assessment and a court ruling before suspension prevents abuse of power while allowing rapid response in emergencies. The company also expressed reservations about laws focusing solely on shutting down models, warning this could create a false sense of security since models executed in customers' isolated environments cannot be remotely withdrawn.

For comparison, OpenAI also wrote to Newsom on October 20, supporting amendments to California's SB 53 to require effective shutdown controls. However, OpenAI's letter did not propose a procedure for the government to seek court orders to pause models. These proposals remain part of the policy debate and have not yet been enacted into state law.

ソース:INSIDE

リサーチ

NXP and Volkswagen Announce Strategic Partnership; S32K5 Chips to Power First Zone-Architecture SDV in 2027

トピック · 恩智浦大众S32K5合作重大
2026-10-29 18:30 GMT+8

Dutch chipmaker NXP Semiconductors and German automaker Volkswagen announced on October 26, 2026, that they have entered into a strategic partnership. Under this agreement, NXP's S32K5 family of automotive-grade microcontrollers (MCUs) will serve as core components in the "zone architecture" of Volkswagen's upcoming "software-defined vehicles" (SDVs). Volkswagen stated that the first SDV based on this architecture is scheduled to be unveiled in 2027.

This collaboration highlights the accelerating shift in the automotive industry from traditional "domain" electronic/electrical architectures to more efficient "zone" architectures. In a zone setup, the vehicle is divided into physical regions (e.g., front-left, rear-right), each managed by a high-performance MCU rather than numerous distributed Electronic Control Units (ECUs). This design aims to reduce wiring complexity, lower weight, and improve data processing efficiency, providing the underlying hardware support necessary for advanced autonomous driving and complex cockpit functions.

Notably, these chips will be used not only in Volkswagen brand models but also in vehicle platforms developed by RV Tech, the joint venture between Volkswagen and US electric vehicle maker Rivian. NXP CEO Rafael Sotomayor described the partnership as contributing to a scalable foundation for next-generation intelligent cars, offering real-time control and edge intelligence capabilities. Karsten Schnake, Volkswagen's board member for purchasing, emphasized that semiconductors have become strategically critical technology for the auto industry, marking this deal as a key step in expanding Volkswagen's semiconductor ecosystem.

ソース:EE Times Japan

リサーチ

Synaptics Announces Astra SR200 Series xMCU with Up to 400GOPS On-Device Multimodal AI Inference

トピック · Astra SR200クイックテイク
2026-10-29 20:30 GMT+8

US semiconductor company Synaptics announced the “Astra SR200” series of extended microcontrollers (xMCU) on October 29, 2026. The product line aims to overcome the memory capacity and computational performance constraints of traditional MCUs by integrating high-bandwidth custom DRAM and dedicated AI accelerators to enable on-device multimodal AI processing.

According to EDN Japan, the Astra SR200 series utilizes Synaptics’ “Torq AI” engine and features a Transformer-compatible “Coral NPU” developed by Google Research, delivering up to 400 GOPS (billion operations per second) of AI inference performance. The architecture combines Arm Cortex-M85 and M52 cores with integrated image and audio processing capabilities. Synaptics states that its hierarchical architecture allows combining up to 8 MB of SRAM with scalable, high-bandwidth custom DRAM, enabling workloads to run at one-third the power consumption of typical system DRAM.

The series consists of three pin-compatible product lines: the SR210 offers 200 GOPS for always-on voice processing and sensing; the SR230 adds image processing and display functions with 300 GOPS; and the SR250 targets advanced edge applications with dual image and display paths, providing 400 GOPS. These specifications are vendor-reported and have not yet been independently verified, nor has a specific mass production delivery timeline been disclosed.

ソース:EDN Japan

リサーチ

Cloud Security Firm Wiz Discloses 'Gheist' Attack: Stolen GitHub Tokens Used to Inject Malicious Workflows in 850+ Accounts, Private Repos Lack Protection

重大
2026-10-29 08:00 GMT+8

US-based cloud security firm Wiz Research recently disclosed an attack campaign named "Gheist," where hackers abused stolen GitHub access tokens to inject malicious GitHub Actions workflows into code repositories, attempting to steal sensitive data. Researchers found that the campaign affected over 850 compromised GitHub accounts, with more than 13,500 malicious branches planted in related repositories.

Wiz has been tracking the Gheist activity since September 15, 2026, and observed signs continuing through October 24. GitHub previously introduced protection mechanisms for public repositories requiring manual approval for certain suspicious workflows, meaning the malicious workflows checked by researchers failed to execute in those contexts. However, this protection does not apply to private repositories or GitHub Enterprise Server. Wiz noted that malicious workflows successfully executed in some enterprise environments.

The injected malicious GitHub Actions workflows encoded accessible environment variables and secrets into logs during execution, bypassing GitHub's existing secret masking mechanisms. The attack tool then called the GitHub API to download these logs and deleted the execution records within seconds to hide traces. Attackers also attempted to search for potentially stored private keys and cloud service authentication credentials from the compromised organizations' self-hosted GitHub Actions runners.

In some enterprise environments, researchers found that attackers used stolen GitHub tokens to unauthorizedly copy private repositories. Hours later, usage records appeared showing queries to AWS Secrets Manager resources using the same organization's cloud service authentication credentials. Wiz researchers told cybersecurity media outlet Dark Reading that they estimated approximately 12 companies were affected by this wave of attacks.

To mitigate such attacks, enterprises should check GitHub repositories for abnormal branches or malicious workflows and compare audit logs between GitHub and cloud services to trace token abuse. If signs are detected, organizations must revoke abused tokens, rotate potentially leaked secret keys and authentication credentials, and inspect self-hosted runners for compromise.

ソース:iThome

リサーチ

South Korean Blockchain Firm DSRV Signs MOU with Ethiopian State-Owned Enterprise FaydaVerse on Digital ID Integration

クイックテイク
2026-10-29 10:00 GMT+8

South Korean blockchain infrastructure company DSRV announced on October 29 that it has signed a Memorandum of Understanding (MOU) with FaydaVerse Digital Solutions Enterprise, a state-owned enterprise in Ethiopia. The two parties will collaborate to integrate blockchain technology into Ethiopia's national digital identity system to build a related financial and public service ecosystem.

FaydaVerse was established this year as a specialized Digital Public Infrastructure (DPI) company spun off from Ethiopia's National Digital ID Program. It operates under the country's sovereign wealth fund, Ethiopian Investment Holdings, and currently provides eKYC, digital wallets, and e-government services based on open-source decentralized identity (DID) technology.

Under the agreement, the companies will form a joint working group to advance four key areas: upgrading the Verifiable Credentials (VC) functionality of FaydaVerse's digital wallet 'FaydaPass'; jointly researching 'FaydaChain', a blockchain infrastructure for institutional digital proofs and qualifications linked to the wallet; developing Web3 financial wallets and technical standards based on identity authentication; and exploring models for distributing electronic vouchers and other public aid via digital IDs.

Jung Ho-sung, Head of Global Business at DSRV, stated that many African countries are increasingly seeking to connect their national digital IDs with financial services and public sectors. He expressed hope that combining DSRV's technology with FaydaVerse's national infrastructure would allow more people to benefit from these technological advancements. Additionally, DSRV is actively negotiating DPI introductions with governments in other major African nations, including Côte d'Ivoire and Senegal.

ソース:zdnet.co.kr

リサーチ

South Korea's SK AX Launches G-Fit Studio AI Tool, Assessing 1,800 Candidates in 6 Hours

クイックテイク
2026-10-29 10:12 GMT+8

South Korean IT services firm SK AX announced on October 29 the launch of "G-Fit Studio," an artificial intelligence competency diagnostic solution designed to assess job seekers' problem-solving abilities and generative AI usage proficiency through simulated real-world work scenarios. This move signals an expansion of AI applications from coding assistance to core human resources selection and development.

The solution employs a two-stage evaluation mechanism: first, candidates interact with an AI agent to solve business problems, with their questioning strategies and decision-making processes recorded; second, the AI acts as an interviewer for a roughly 30-minute deep-dive session to test logical thinking and collaboration skills. SK AX reported that the tool was applied to the fourth cohort of its "SKALA" talent program, completing capability measurements for over 1,800 applicants in just six hours, significantly reducing the operational burden of large-scale recruitment.

Beyond hiring, SK AX plans to utilize the tool for existing employees' skill diagnostics and customized training, aiming to build an AI-managed system covering the entire talent lifecycle. Kim Min-hwan, head of SK AX's HRX promotion team, stated that corporate competitiveness depends on effectively acquiring and nurturing talent with practical AI capabilities. Currently, there is no third-party independent data supporting the accuracy or bias mitigation of the product's assessments.

ソース:zdnet.co.kr

リサーチ

Malaysia Customs Seize $12 Million in AI Chips Suspected for China Smuggling Network

トピック · 马来西亚AI芯片走私案重大
2026-10-29 01:02 GMT+8

According to Reuters, Malaysian Royal Customs officials seized 30 servers containing over $12 million (approximately RM49.5 million) worth of advanced AI chips at the Kuala Lumpur International Airport free trade zone on October 15, 2026. The goods were falsely declared as ordinary electronic products to evade export restrictions.

Kuala Lumpur International Airport Customs Director Aspalila Ag Tuah confirmed that the suspected smuggling network used Malaysia as a transit point to bypass restrictions on the onward movement of the goods. Investigators suspect this operation is linked to a similar case from June last year, when customs discovered 72 server units containing advanced AI chips valued at approximately $13 million at the same location. Aspalila stated that the network has been linked to seizures worth over $25 million across at least five cases.

While specific details such as chip type, country of manufacture, final destination, and intended recipient remain undisclosed, analysts suggest China is the most likely destination. Due to U.S. export controls on advanced Nvidia and other U.S.-origin AI chips destined for China, Malaysia—a U.S. ally not subject to these direct restrictions—has become an ideal transit hub for smuggling sensitive technology, leveraging its mature logistics infrastructure. Previously, at Washington's request, Malaysia imposed export controls on the movement of advanced AI chips of U.S. origin.

ソース:Tom's Hardware

リサーチ

Silicon Valley 101 Interviews Ethereum Co-founder Vitalik Buterin: Reflecting on Hundreds of Millions in AI Safety Donations and d/acc Philosophy

重大
2026-10-29 07:00 GMT+8

The podcast Silicon Valley 101 released Episode 257, titled "Dialogue with Vitalik: Philosophical Thoughts Behind AI Safety and How Effective Altruism Influences Silicon Valley," on October 28, 2026. The episode features Ethereum co-founder Vitalik Buterin and tech journalist Afra, focusing on the tenets of Effective Altruism (EA) and rationalist communities, and how these ideologies shape current AI safety research directions and funding allocations.

According to the publisher's notes, Vitalik Buterin previously donated substantial funds to institutions within the EA circle. In the program, he recalls how a cryptocurrency donation unexpectedly grew into hundreds of millions of dollars, and discusses why the subsequent direction taken by the recipient organizations prompted him to re-evaluate his initial choices. This first-person account of fund flows offers a unique perspective on the interest networks and power structures underlying the AI safety field.

Beyond reviewing donation history, the episode explores Vitalik's concept of "d/acc" (defensive accelerationism). Unlike e/acc, which advocates accelerating technological development, or PauseAI, which calls for halting frontier development, d/acc prioritizes developing technologies that protect individuals and decentralize power—such as cybersecurity and vaccines—to enhance society's resilience against risks. The discussion also covers controversies surrounding evaluation bodies like METR, the balance between commercial interests and safety at companies like Anthropic, and how ordinary people can understand debates around AI uncertainty.

The entire episode is recorded in Chinese, with Afra providing background context in the first half and the hosts conversing with Vitalik in the second half. Given the numerous English names and institutions involved, the publisher has compiled a detailed glossary in the show notes and released subtitled versions on platforms like Bilibili and YouTube for easier reference.

ソース:sv101.fireside.fm

リサーチ

Samsung Plans Preferred Share Buybacks for Remaining Returns to Bypass Financial Affiliate Ownership Limits

トピック · 三星电子2026资本支出重大
2026-10-29 09:29 GMT+8

South Korean tech giant Samsung Electronics disclosed during its third-quarter earnings conference call on October 29 that its remaining shareholder return funds for this year amount to between 55 trillion and 75 trillion KRW. This figure represents the balance after deducting the 35 trillion KRW already executed or finalized from the total annual shareholder return plan of 90 trillion to 110 trillion KRW.

Samsung Electronics CFO Park Soon-chul stated that the company recognizes the market's high interest in the price difference between common and preferred shares. He added that when repurchasing and cancelling treasury stock, the company will actively consider and adjust the proportion of preferred share purchases. This indicates that a substantial part of the remaining funds will be used for preferred share buybacks and cancellations rather than traditional cash dividends or common share cancellations.

This strategic shift is directly linked to ownership restrictions on financial affiliates under South Korea's Financial Industry Structure Improvement Act. Currently, Samsung Group's financial subsidiaries, Samsung Life Insurance and Samsung Fire & Marine Insurance, hold approximately 10% of Samsung Electronics' common shares combined, reaching the regulatory limit. Large-scale cancellation of common shares would passively increase these financial affiliates' relative ownership ratios, potentially triggering additional divestiture obligations. In contrast, cancelling non-voting preferred shares does not affect the voting rights distribution of common shares, allowing capital returns without violating regulatory red lines.

Additionally, the Samsung Electronics board resolved on the same day to pay a third-quarter cash dividend of 4,601 KRW per share for both common and preferred stocks, totaling approximately 30 trillion KRW, with payment expected in November. The specific execution plan for the remaining shareholder return funds and the new shareholder return policy post-2027 are scheduled to be finalized and announced at a board meeting in late January next year.

ソース:미래를 보는 창 - 전자신문

リサーチ
次へ →
ログイン / 新規登録
読書リサーチレーダー投資フレームワーク

They implement WYSIWYG using CodeMirror 6 decorations and widgets: headings appear larger, markup symbols are hidden, tables are drawn as tables, and formulas are rendered as formulas, while the underlying text remains unchanged. Three views—the document view, live preview, and source view—are different presentations on the same CodeMirror instance. Switching between them requires no conversion, preserving selection ranges and undo history.

Regarding performance, they provide proof of reuse for Lezer's Markdown incremental parsing. The parser records which intervals are reused verbatim, allowing the editor to skip unchanged parts. For a 1 MiB document with continuous input over 20 iterations, syntax tree visits dropped from 107,760 to 1,380 (a 98.7% reduction); syntax tree comparisons dropped from 153,940 to 660 (a 99.6% reduction). Where proofs fail, the system falls back to full comparison, ensuring correctness is unaffected.

The author explicitly acknowledges the costs of source-first approaches: every input requires updating the syntax tree, making large files slow. MSFE is currently in alpha, and its API is subject to change. The open-source portion is licensed under MIT; commercial components include AI document reading/writing, displaying changes and annotations in a 'pending review' format. They also promote the QuireKit editor ecosystem, including Classic, a page-based document editor based on ProseMirror.

Notable boundaries exist: these performance figures are self-reported benchmarks. The portability of the code fence solution for complex blocks—i.e., whether other Markdown tools can recognize and edit structured data within fences—has not been verified by third parties. The author's core argument (that tree-based authority leads to source loss) is built on reading specific library versions' source code and is verifiable; however, the long-term maintenance costs and ecosystem compatibility of the alternative solution remain open questions.

This argument has its boundaries. The 47% gap under the exchange rate metric is itself influenced by exchange rate cycles: the figure widens when the US dollar strengthens and narrows when the renminbi appreciates. Furthermore, exchange rate fluctuations are continuous and observable in real-time, whereas ICP revisions are discrete and occur in concentrated bursts only once every three to six years; there is a structural difference in frequency and predictability between the two. Even though the revision magnitude is larger, whether policymakers are willing to accept a baseline dataset updated only once every few years versus a sequence of exchange rates observable daily remains a distinct practical issue. Frankel's argument holds logically—that excluding exchange rates due to volatility while accepting the even more volatile PPP is inconsistent—but readers should note that this conclusion relies on directly comparing the magnitudes of two fundamentally different types of instability.