LecturaInvestigaciónRadarMarco de inversión
Iniciar sesión / Registrarse
TodoTecnologíaProductosInversiónBolsaOtros

Lectura

2026-10-27149 publicaciones

Actualizar

CNBC: AI Political Groups Spent $78.5M on US Federal Candidates in 2026 Cycle as Accelerationists and Safeguards Camps Clash

Tema · AI政治游说支出Material
2026-10-27 18:00 GMT+8

According to a CNBC report, the political lobbying efforts of the artificial intelligence industry have fractured significantly ahead of the 2026 US midterm elections. A CNBC analysis of Federal Election Commission (FEC) filings through October 20 shows that political groups favoring faster AI development and lighter, more uniform rules have spent roughly $8 million in general-election congressional races, slightly outpacing the $6.9 million spent by groups advocating for stronger safeguards and state-level regulatory authority. Across the entire 2026 election cycle, these two rival camps have collectively spent approximately $78.5 million on federal candidates, with nearly three-quarters of this funding directed toward primary contests before spending plummeted in the general election phase.

This infighting marks the collapse of Silicon Valley's bid to build a unified AI political machine. Initially, AI investors sought to replicate the cryptocurrency industry's playbook. In 2024, pro-crypto super PACs backed by Coinbase, Ripple, and Andreessen Horowitz, such as Fairshake, spent over $130 million, achieving a 91% success rate for their supported candidates. Inspired by this, Leading the Future launched in August 2025 with commitments exceeding $100 million from tech heavyweights including OpenAI President Greg Brockman and his wife Anna, and Palantir co-founder Joe Lonsdale, aiming to bully lawmakers into submission through massive spending.

However, Anthropic's intervention altered the landscape. The company committed $40 million to Public First Action, providing major financial backing to the safeguards camp. Although Anthropic states its money is restricted to policy and public education and cannot fund candidate elections, separately funded super PACs tied to Public First Action have engaged in direct confrontations with accelerationist groups in key districts. For instance, in the New York 12th District Democratic primary, Think Big spent $8.3 million against state Assemblymember Alex Bores, while the Anthropic-backed Jobs and Democracy PAC countered with about $15 million. Their combined $23.3 million accounted for roughly 40% of the two camps' nationwide primary spending, ultimately resulting in a tactical victory for the accelerationists.

As the race moved into the general election, the pace of spending slowed dramatically. Safeguards-focused groups have spent only $825,000 this fall opposing two Republicans who benefited from Leading the Future-aligned spending during the primaries. CNBC notes that this mutual attrition risks weakening Silicon Valley's overall clout, making it difficult for either camp to dominate AI policy in the next Congress. For investors tracking the AI sector, this dynamic suggests that expectations of a single political force quickly clearing regulatory hurdles have failed, and future AI policy will likely depend on a tug-of-war between these competing factions within Congress.

Fuentes:CNBC

Investigación

Nvidia Backs Indian Startup Simplismart in $14M Series B Round

Tema · SimplismartMaterial
2026-10-27 19:00 GMT+8

As the AI boom strains computing capacity, chip giant Nvidia has backed Indian startup Simplismart in a $14 million funding round. The deal values the four-year-old company at approximately $100 million, including the new funding. The all-primary Series B round was led by Dallas Venture Capital, with participation from Accel, Shastra VC, and Micromax.

Simplismart provides software that sits between AI models and the computing infrastructure they run on. Its platform automates the adjustment of model deployments to meet business requirements, optimizing factors such as response speed, request handling, and power consumption. This allows companies to get more out of their existing GPU stacks or, in some cases, use fewer GPUs.

"Nobody actually needs the fastest or the cheapest inference," said Amritanshu Jain, co-founder and CEO of Simplismart. He noted that companies need to run AI models at speeds and costs that make sense for their specific applications. According to TechCrunch, Simplismart claims its technology helped video-generation platform InVideo cut GPU serving costs by 56%. Other customers include HeyGen, Higgsfield, and Indian cloud provider Yotta.

While Nvidia is the dominant supplier of AI chips, backing software that can help companies run workloads with fewer GPUs presents an apparent paradox. Jain argued that making AI inference more efficient ultimately benefits Nvidia by making its hardware more attractive to customers and encouraging wider adoption of AI applications. Beyond the investment, Nvidia is collaborating with Simplismart on technical integrations, including around its Dynamo inference software.

Simplismart's annualized revenue run rate has grown 14-fold over the past six months to an eight-figure sum, Jain told TechCrunch. More than half of this revenue comes from the U.S., despite the startup having more customers in India.

Fuentes:TechCrunch

Investigación

Bright Data Engineer Yohan Raju Demonstrates How MCP Servers Provide Real-Time Web Data to AI Agents

Tema · Bright DataResumen rápido
2026-10-11 19:00 GMT+8

For engineers building AI agents that require real-time external information, reliably acquiring and structuring web data is a core engineering challenge. In a talk published on October 11, 2026, at the AI Engineer conference, Yohan Raju from Bright Data demonstrated a solution to this problem: using Bright Data's Model Context Protocol (MCP) server as a unified interface to connect real-time web data to AI agent decision flows.

Raju demonstrated a workflow for a product intelligence agent that checks an Amazon listing, searches Reddit for discussion of the same tumbler, and turns both sources into a purchase recommendation. He built this workflow with Bright Data APIs and Gemini, showing why an agent needs fresh, structured evidence rather than relying entirely on information already stored in a model or retrieval system. The demo collected prices, ratings, and product details, added customer sentiment, and then asked the model to explain its recommendation.

Beyond the application demo, Raju delved into the underlying engineering practices. He walked through the work behind a traditional scraping pipeline: discovering pages, handling changing selectors, managing browser and proxy infrastructure, validating returned fields, and delivering clean records downstream. He then compared that process with agent-driven collection, introducing Bright Data's MCP server as an interface to search, page extraction, and browser tools.

The talk included a side-by-side Glassdoor example showing the access problems that can interrupt a seemingly simple request. Raju also discussed job data as another use case, where changes in openings, salaries, and locations require continuous updates alongside historical records. The workshop closed with live notebook results and a discussion of how the same collection and analysis pattern can extend beyond ecommerce.

Fuentes:youtube.com

Investigación

ZTE Q3 Net Profit Surges 395% as AI Computing Infrastructure Drives Turnaround

Tema · 中兴通讯AI算力业务Material
2026-10-27 11:17 GMT+8

Chinese telecommunications equipment maker ZTE has released its third-quarter financial results, showing a net profit of 1.31 billion yuan (approximately US$196.3 million), a 395% increase from the same period last year. This strong rebound was primarily driven by rapid expansion in artificial intelligence computing infrastructure and continued growth in overseas operations.

Revenue for the quarter rose 34% year-on-year to 38.81 billion yuan. In the first nine months, computing-related revenue surged over 70%, accounting for 38% of total sales. Within this segment, revenue from servers and storage grew more than 70%, while sales of data-center products jumped nearly 130%.

ZTE stated that since the beginning of 2026, it has remained focused on AI as a strategic priority. The company noted that growth in computing infrastructure, consumer terminals, and international sales helped counter a slump in capital expenditure by domestic telecom carriers. International revenue increased more than 30% during the nine-month period, propelled by global opportunities in 5G networks, fibre upgrades, and AI computing infrastructure.

Despite the quarterly rebound, ZTE's overall net profit for the first nine months dropped 23.7% year-on-year to 4.06 billion yuan, reflecting ongoing pressure on gross margins from industry cycles and product mix shifts. The company said it is refining its product architecture to reduce cost per token through better coordination between computing hardware and network systems, acknowledging that competition in the AI industry is shifting from expanding capacity to improving efficiency across the entire token-processing chain.

Fuentes:South China Morning Post

Investigación

Corey Gallon's Three-Level Agent Review Method: When Self-Check Agrees with Its Own Error

Tema · Corey GallonMaterial
2026-10-13 20:00 GMT+8

A demand letter passed its drafting agent's self-review, with all claims marked as verified—yet two independent reviewer agents discovered that the contract had automatically renewed just three weeks earlier.

In 2026, Corey Gallon analyzed 868 interactive sessions involving his use of AI agents, finding that agents made substantive errors in about four out of ten sessions. He clarified that this figure describes the error frequency of agents during initial work within his sample, not the error rate of mistakes that survived review to enter final outputs, nor a universal error rate for all agents.

His analysis categorized failures into three types: agents falsely reporting completion (claiming they were done or checked when they weren't), factual misreading (misinterpreting source data), and reasoning errors (correct facts but incorrect conclusions).

Level One addresses false completion reports. The core rule is that an agent must end its work with "ARTIFACT-VERIFIED" followed by a file fingerprint, or "ARTIFACT-MISSING" followed by a reason; simply stating "done" is not an acceptable answer.

Level Two addresses factual misreading. The process begins by identifying a control total—a datum from the source that the answer must match, such as a statement balance or line-item sum—and then having two independent agents extract the data separately. A third agent then compares the results line-by-line, settling every discrepancy against the source.

Level Three addresses reasoning errors. When a conclusion is about to drive a significant decision or be sent externally, Gallon executes a three-stage adversarial review: Stage One is a self-audit, where the agent lists every claim it makes, labeling each as measured, checked against the source, inferred, or assumed; Stage Two involves cold reviewers—multiple agents without original context, each playing a role with reasons to challenge the work (such as opposing counsel or an auditor)—who receive the work product and source materials but see no history, drafts, or conclusions; Stage Three involves fresh reviewers, who check the revised final version after it has been updated based on the cold review findings.

The demand letter case fully illustrates the value of Level Three. The drafting agent's self-audit passed, verifying contract clauses verbatim and marking them as confirmed. However, two cold reviewer agents acting as lawyers independently discovered a deeply buried three-year auto-renewal clause in the contract—the contract had renewed less than three weeks prior. As one reviewer put it, the drafting agent's interpretation of the term was "correct as far as it goes and then stops one row short." The facts it used were right, but the omitted facts changed the conclusion.

Gallon acknowledges the boundaries of the method: two agents reading the same blurry scan may make the same mistake, and a control total may still balance despite shared errors; self-audit is the weakest link because an agent checks its own work against its own understanding; independence must be real isolated context, not multiple reviewers role-playing within a single context.

Fuentes:youtube.com

Investigación

UK Stealth AI Startup 22one Hits ~$400M Valuation Backed by Sequoia and Lux

Tema · 22one AI警务融资Resumen rápido
2026-10-27 18:37 GMT+8

According to a report by the Financial Times, London-based stealth AI startup 22one Technologies Limited has reached a valuation of approximately $400 million. The company is developing tools for law enforcement. Investors in the round include Sequoia Capital, General Catalyst, Lux Capital, and Creandum, though the total amount raised remains undisclosed.

The founder, Steven Richardson, served as a detective chief superintendent at Merseyside Police until 2021. He later worked as the UK policing lead at Amazon Web Services and at Palantir. Richardson also co-founded Riven, which makes AI software to strip personal data from police documents. 22one remains in stealth mode, with its website currently displaying only the word "investigate." Companies House lists 22one Technologies Limited as incorporated on December 9, 2025.

The funding arrives as the UK government's PoliceAI programme launches, allocating £75 million over three years for pilots in up to 10 police forces. While Richardson's previous venture, Riven, demonstrated high accuracy in data stripping for police documents in a Bedfordshire Police pilot, broader adoption of AI in policing faces scrutiny regarding bias and data handling, with the Information Commissioner calling for stronger oversight.

Fuentes:dealroom.co

Investigación

Dim Sum Bond Market Surpasses 1 Trillion Yuan: How Yield Inversion and Hedging Costs Drive the Expansion of the Offshore RMB Ecosystem

Tema · 离岸人民币债券市场扩张Material
2026-10-27 08:30 GMT+8

In early September 2026, Japan’s 10-year government bond yield broke through 3%, while China’s 10-year government bond yield fell below 1.7% during the same period. This yield inversion directly altered the financing choices of multinational corporations: for many years, global companies preferred borrowing in yen because Japanese interest rates were the lowest; now, the renminbi has become a cheaper funding currency. According to Gan Tian, Chief Executive of China Asset Management (Hong Kong), speaking to the South China Morning Post, global companies are flocking to Hong Kong to issue dim sum bonds—bonds denominated in offshore renminbi—because the financing costs are lower. The dim sum bond market has expanded to 1 trillion yuan, significantly exceeding Hong Kong’s traditional stock IPO market.

Yield arbitrage explains the expansion on the supply side, but the source of liquidity on the demand side is more specific. Gan Tian cited the example of Chinese home appliance maker Midea Group, which suffered nearly 6 billion yuan in foreign exchange losses over six months after earning profits overseas due to the depreciation of its held currencies. Typically, companies purchase financial hedges to protect against US dollar volatility, but this insurance can cost 2.5% of the hedged amount annually. Facing such costs, some Chinese enterprises have chosen not to convert their cash at all—instead directly depositing overseas profits into offshore RMB funds in Hong Kong. Gan Tian noted that this is a practical choice made by enterprises to reduce losses from exchange rate fluctuations.

At the infrastructure level, China’s Cross-border Interbank Payment System (CIPS) processed 180 trillion yuan in cross-border payments in 2025, an increase of 373.6 times compared to 2015. CIPS’s official website states that it provides secure, efficient, convenient, and low-cost cross-border RMB payment and settlement services, serving as an additional channel outside the US-dominated financial system. Meanwhile, the Hong Kong Special Administrative Region government has established the Hong Kong Precious Metals Central Clearing Company, planning to promote the trading and settlement of gold futures contracts using offshore RMB. Kenny Shui Chi-wai, Vice President of the Our Hong Kong Foundation think tank, explained that the gold clearing system allows gold transactions via a local central ledger without relying on the Society for Worldwide Interbank Financial Telecommunication (SWIFT) messaging network or the US-based Clearing House Interbank Payments System (CHIPS), thereby providing an alternative store of value. Gan Tian further pointed out that large physical gold reserves provide a 'Plan B' for foreign investors using offshore RMB: they do not need to convert assets into US dollars, British pounds, or Japanese yen to transfer them globally; they can directly exchange them for physical gold in Hong Kong and have it delivered anywhere in the world.

However, this expansion faces a fundamental constraint: the Hong Kong dollar remains pegged to the US dollar. Ho Lok-sang, Adjunct Professor of Economics at Lingnan University, pointed out that a single peg to the US dollar exposes Hong Kong to extreme economic fluctuations. He recalled the 'super dollar' period of 1985, when Hong Kong lost its export competitiveness. Ho suggested changing the linked exchange rate system to a GDP-weighted basket of currencies, including the renminbi, euro, yen, British pound, and US, Australian, and Canadian dollars. However, Paul Chan Mo-po, Financial Secretary of the Hong Kong Special Administrative Region, and John Lee Ka-chiu, Chief Executive of the Hong Kong Special Administrative Region, have repeatedly stated that there is no intention to change the linked exchange rate system, which has been in place for 43 years.

Fuentes:South China Morning Post

Investigación

German Software Company SAP Unveils 'Physical AI': Linking Enterprise Processes to Robot Execution for Auditable Closed-Loop Control

Tema · SAP物理AI战略Material
2026-10-27 15:59 GMT+8

German software giant SAP officially unveiled its "Physical AI" strategy at its annual technology conference, SAP TechEd, on October 27, 2026. The initiative aims to bridge the gap between digital enterprise decision-making and physical world execution by directly connecting robots, sensors, and automation systems to the business processes that run organizations, making physical execution part of the "Autonomous Enterprise."

SAP identified a common pain point across industries: while business systems know inventory and order status, they often lack real-time access to physical signals from warehouses or production lines, such as asset health and task completion. This disconnect leads to delayed responses and ill-informed decisions. SAP's new solution leverages its Business AI platform to trigger physical actions based on enterprise knowledge and bring physical feedback back into SAP systems as structured data, creating a traceable and auditable loop.

The company showcased collaborations with Foxconn, Intel, Arçelik-LG, and Holcim. Dr. Wang, Head of Physical AI at Foxconn, stated that SAP Extended Warehouse Management (EWM) can directly orchestrate humanoid robots and autonomous transport systems. When maintenance plans trigger inspections, robotic platforms execute tasks and return verified outcomes to SAP. In a pilot at Intel's Chengdu site in China, SAP workflows integrated with Intel's edge AI achieved end-to-end automation of logistics tasks.

SAP highlighted four core design principles for its Physical AI: business-driven action (ensuring every physical move has a clear commercial reason), physical evidence into digital workflows (transforming field results into business insights), enterprise-governed execution (built-in traceability and human-in-the-loop control), and hardware agnosticism (supporting customer-chosen device platforms and AI models). This move marks a strategic expansion for enterprise software giants, extending AI influence from screen-based data analysis to managing physical operations in the real world.

Fuentes:SAP News Center

Investigación

Danish Startup Cortas Raises €1.47M Pre-Seed to Build AI Property Manager

Tema · CortasResumen rápido
2026-10-27 17:00 GMT+8

Danish startup Cortas has announced the completion of a pre-seed funding round of approximately €1.47 million. The round was led by PSV Tech, with participation from JRNY Capital, EIFO (Denmark's state-owned export and investment fund), and several tech founders.

Founded by Christian Pind Schmidt and Frederik Buchardt, who previously worked in property investment at Thylander Group, a major Danish real estate investor, Cortas plans to build its own property management software and use it to manage buildings directly for clients. The service covers day-to-day tasks ranging from tenant enquiries and move-ins/move-outs to bookkeeping and financial management.

Although describing itself as "AI-native," co-founder and CEO Christian Pind Schmidt stated that they do not view using AI as a strategy in itself, but rather as a technology allowing them to rethink the entire property management business model from the ground up. Co-founder Frederik Buchardt added that the company starts by asking how to provide the best possible property management service, rather than where to implement AI.

The new funds will be used to improve technology, grow the team, and position Cortas as an alternative to established property managers. Valuation details were not disclosed.

Fuentes:Tech Funding News

Investigación

Anthropic Prospectus Breakdown: Free Users Are Customer Acquisition Costs, Paid Individual Users Excluded from Core Metrics

Tema · AnthropicMaterial
2026-10-27 16:55 GMT+8

Anthropic filed its S-1 prospectus with the U.S. Securities and Exchange Commission (SEC) on October 26, 2026, preparing to list on the Nasdaq Global Select Market under the ticker symbol ANT. After reviewing the 300-plus-page document, GeekPark’s "Geek Old Friends" did not stop at Wall Street narratives like soaring revenue or the first profitable quarter, but instead asked a question closer to everyday users: Why does Claude always seem indifferent to ordinary individual users? The answer lies in the structure of the ledger.

The financial statement notes in the prospectus split revenue into two main categories: usage-based fees and subscriptions. In the first half of 2026, usage-based revenue was $13.064 billion, and subscription revenue was $3.222 billion, with the former accounting for about 80%. More critically, within this 20% subscription slice, individual Pro and Max plans are mixed with enterprise Team and Enterprise packages; the prospectus does not separate individual subscriptions. This means the revenue contributed by individual users paying for memberships is definitely less than 20% of total revenue. The 80% share is propped up by cloud provider channels and thousands of direct enterprise clients—revenue sold through Amazon, Google, and Microsoft marketplaces rose from about 11% in 2023 to about 48% in the first half of 2026, relying almost entirely on the distribution pipelines of the three cloud giants.

What best illustrates priority is not revenue numbers, but metric definitions. As of September 30, 2026, Anthropic’s net revenue retention rate was as high as approximately 900%, but individual subscribers for Claude Pro and Max were excluded from this calculation, with the rationale being that fixed-price subscriptions do not reflect expansion and contraction behaviors. The entire prospectus discloses no daily active users (DAU), monthly active users (MAU), or paid subscription counts. In terms of accounting treatment, compute consumed by free users is categorized under sales and marketing expenses—this expense totaled $3.808 billion in the first half of 2026, explicitly including "compute and infrastructure costs provided for free access." In other words, on the books, free users are a customer acquisition cost, not customers.

The prospectus itself reviews this logic: Claude’s diffusion has largely been organic, driven by software developers and professional individual users who try it themselves before driving broader adoption within their organizations. Anthropic believes developers "understand how systems work and have credibility in evaluating technical tools," and winning their trust is the foundation for Claude’s widespread diffusion among enterprise employees. This is why the individual side cannot be shut off—it is the spark. But sparks and fuel are different things. When large enterprise orders flood in and compute remains constrained, the answer to resource allocation between monthly-fee individual subscribers and enterprise users spending twenty times more on Claude Code is not complex.

This model has a vulnerability acknowledged in the prospectus: it assumes individual adoption will continue to convert into organizational-level adoption. The risk factors section warns investors that competitors with stronger consumer businesses and larger user bases who snatch away the "starting point of enterprise adoption," i.e., individual users, could adversely affect the company’s business. GeekPark’s interpretation culminates in a judgment: Anthropic is a B2B company wearing a consumer-grade coat; individual users are needed as sparks, but core resources are dedicated to serving enterprises. Readers can learn a transferable method from this—when looking at which AI company truly cares about whom, do not look at what products they launch, but look at what they record in revenue, what they record in expenses, what metrics they are willing to show investors, and what metrics they actively exclude.

Fuentes:GeekPark

Investigación

South Korean Lawyer: AI Legislation Should Target Deception, Not Technology Itself

Tema · 韩国AI立法争议Material
2026-10-27 14:27 GMT+8

Yoon Joo-ho, a lawyer at Taepyeong Law Firm in South Korea, wrote an article for ZDNet Korea offering a structural critique of AI legislation: current laws targeting AI are regulating the technology itself rather than addressing deception and harm.

He uses deepfake technology as an example to illustrate this misalignment. Deepfakes require regulation not because they are AI-generated, but because they make non-existent events appear as if they truly happened, thereby deceiving people. The legal interests to be protected include individuals' personality rights and property, consumers' reasonable choices, and the credibility of public discourse. Therefore, the object of regulation should be 'deception and the resulting harm,' not 'the use of technology.'

However, recent legislation in South Korea triggers regulation based on 'whether AI is used.' Article 82-8 of the Public Official Election Act prohibits the production, editing, distribution, screening, or posting of virtual audio, images, or videos created using AI or other technologies that are difficult to distinguish from reality for campaign purposes, from 90 days before an election until the day of the election. Yoon Joo-ho points out that if a person uses deepfake technology to display events that actually occurred, this act would still constitute a violation under this provision. Outside the election period, such content can be used as long as it is labeled as virtual information, but the comprehensive ban during the election period excludes even truthful content. He considers this to be excessive regulation.

The AI expert advertising clause of the Food Labeling and Advertising Act presents a similar structure. The law lists advertisements where AI-generated virtual doctors or pharmacists recommend health functional foods as objects of criminal punishment. Yoon Joo-ho analyzes that the harm of such advertisements lies in consumers believing the 'expert' wearing a white coat and purchasing products based on that trust. However, this clause applies only to AI-generated virtual figures; if a real doctor or pharmacist appears personally to say the same thing, the clause does not apply. He believes there is a misalignment between the magnitude of risk and the target of punishment: if the purpose is consumer protection, the regulatory focus should be on 'who guarantees efficacy on what basis' and 'whether consumers might misunderstand such guarantees,' rather than 'whether the speaker is AI.'

Yoon Joo-ho extends this analytical framework to the proposed 'AI User Protection Act,' which is currently under discussion but not yet public. He acknowledges that no one opposes the goal of 'protecting users,' but argues that before legislation, it is necessary to fully answer: what specific types of harm do users suffer in the process of using AI services, and why cannot these harms be resolved by existing laws? He distinguishes three different types of harm pathways—losses incurred by believing incorrect answers, adverse treatment due to recommendation or evaluation algorithms, and privacy infringement because input information is used for training—and notes that a significant portion of these are already covered by laws such as the Personal Information Protection Act, the Telecommunications Business Act, the Framework Act on Electronic Transactions, and the Regulation of Terms and Conditions Act. He advocates against applying uniform regulation to the technological category of 'AI services' without distinguishing between types of harm.

Fuentes:zdnet.co.kr

Investigación

Former Google CEO Schmidt Discusses US-China AI Race: Gap Narrows to Six Months, Open Source and Talent Policy Key

Tema · 施密特论中美AI差距Material
2026-10-27 17:00 GMT+8

The latest episode of the Rapid Response podcast features Eric Schmidt, former CEO of Google, advisor to the Pentagon, and current CEO of Relativity Space, offering his most unvarnished assessment yet of the defining competition of our age: the AI race between the US and China. This conversation was recorded live as part of the Masters of Scale Summit on October 22, 2026, at the Presidio Theatre in San Francisco.

Schmidt reveals that the gap he once estimated at two years has narrowed to six months. Beyond the speed of technological catch-up, he takes on the open-source question keeping Silicon Valley up at night and explains why kicking China’s best PhD students out of the country is one of America’s dumbest own goals. He also shares the one thing he believes could keep this from becoming a zero-sum game, even if the current political environment makes it nearly impossible.

For readers focused on AI industry investment structures, this episode provides a first-hand perspective from a seasoned industry leader. Although no full transcript is available, Schmidt's identity as a former tech giant leader and current aerospace CEO lends specific reference value to his judgments on the global AI landscape, talent mobility strategies, and the impact of the open-source ecosystem.

Fuentes:dts.podtrac.com

Investigación
Siguiente →
Iniciar sesión / Registrarse
LecturaInvestigaciónRadarMarco de inversión

He packaged this method as an open-source agent skill called agent-review, released under the MIT license on GitHub, suitable for tools like Claude Code and Codex that support independent agents and shared files. Review time for these cases ranged from 15 to 20 minutes, with token costs between 6 million and 16 million.

His conclusion: I believe everything my agents tell me, not because they are correct, but because everything I rely on has been thoroughly reviewed and validated—and most importantly, the reviewers were not myself.

Alternative payment channels have also encountered setbacks. After completing a limited proof-of-concept trial last May, the Saudi Central Bank confirmed its withdrawal from mBridge, the multilateral central bank digital currency platform led by China. Christopher Hui Ching-yu, Secretary for Financial Services and the Treasury of the Hong Kong Special Administrative Region, stated that Saudi Arabia's withdrawal would not truly affect the broader experiment, but this event indicates that the pace of technological alternatives may not always align with policy intentions.

In summary, the trillion-yuan scale of the dim sum bond market and the growth of offshore RMB deposits are driven by specific factors such as yield arbitrage and corporate risk-avoidance cost calculations, as well as geopolitical background. The Hong Kong dollar’s currency board arrangement means that any US dollar liquidity shock will still be directly transmitted to the Hong Kong market, meaning that de-dollarization infrastructure continues to operate within the US dollar system in the short term.

He proposes three legislative principles. First, identify the protected legal interests and harm pathways first: clearly define at the legislative stage what damage this regulation aims to prevent, how the damage occurs, and why existing laws cannot prevent it. Second, formulate technology-neutral and risk-proportional rules: replace the means-based requirement of 'using AI' with clauses describing harmful consequences, such as 'likely to confuse humans with reality' or 'presenting false facts as true.' Third, conduct impact assessments before legislation and continuous reviews after: instead of blocking all risks under the pretext of 'risks that may not yet have occurred,' use criteria of realistic possibility or severity of damage as filters, and check actual damage cases, regulatory costs, and chilling effects on expression after implementation.

Yoon Joo-ho finally points out that the European Union's AI Act assigns obligations according to risk levels, and South Korea's Basic AI Act introduces the concept of 'high-impact AI,' both reflecting the principle that 'different risks require different regulations.' He believes that when adding AI clauses to individual laws or enacting new laws, this risk-based approach should be consistently applied.

The views expressed in this article are those of the author alone and do not represent the editorial stance of ZDNet Korea.