LecturaInvestigaciónRadarMarco de inversión
Iniciar sesión / Registrarse
TodoTecnologíaProductosInversiónBolsaOtros

Lectura

2026-10-28154 publicaciones

Actualizar

Arize AI Engineer Fuad Ali Demonstrates Self-Improving Loop for AI Agents: From Trace Logs to Automated Deduplication and Evals

Tema · Arize AIResumen rápido
2026-10-27 07:00 GMT+8

Fuad Ali of Arize AI presented a workshop video on the AI Engineer channel demonstrating a self-improvement loop architecture for AI agents. This approach aims to address the difficulty of automatically identifying and fixing defects in long-running agent systems, providing concrete operational reference for engineering teams.

According to the publisher's chapter list and description, Ali details the key steps of this closed loop: first, comprehensive tracing of agent actions; second, an always-on agent named Signal reads these trace logs on a schedule and automatically files deduplicated issues; third, detected failures are converted into evaluations (evals) and living regression datasets; finally, experiments are run to prove that fixes work.

Within this automated process, Ali specifically notes that human engineers should concentrate on the highest-leverage points, including eval design, dataset curation, and reviewing fixes. The video also includes a live demo on a toy-shopping agent with a planted bug, along with Q&A sessions on evaluating qualitative output, decision models, and cost.

For engineers building or optimizing autonomous AI agent systems, this resource offers a concrete example of a workflow from log monitoring to automated evaluation feedback, helping to understand how to transform manual debugging into a semi-automated continuous improvement mechanism.

Fuentes:youtube.com

Investigación

General Motors Announces Move of Stock Listing from NYSE to Nasdaq Effective Nov. 17

Tema · 通用汽车转板纳斯达克Material
2026-10-28 07:54 GMT+8

According to Bloomberg, U.S. automaker General Motors (GM) announced on October 27 that it will move its stock trading from the New York Stock Exchange (NYSE), where it has been listed for 115 years, to the Nasdaq Global Select Market effective November 17. The ticker symbol 'GM' will remain unchanged.

The migration is part of a strategic repositioning. GM aims to shed its traditional automaker image by highlighting investments in autonomous driving, electric vehicles, and software, presenting itself to investors as a technology growth company. GM Chief Financial Officer Paul Jacobson stated that Nasdaq is a better fit for the company's technology-centric growth strategy.

Market attention focuses on potential inflows from index-tracking funds. JPMorgan analysts noted that if GM meets eligibility criteria in December, it could be added to the Nasdaq-100 Index. By market cap, GM is currently the second-largest company not yet included in the index, behind Hilton. However, index providers have not confirmed this potential addition.

GM management has previously argued that the stock price undervalues the company despite strong cash generation. Earlier this month, the company announced an additional $20 billion share repurchase plan. Following the listing change announcement, GM shares rose approximately 4% around 9:33 a.m. New York time.

Fuentes:zdnet.co.kr

Investigación

Amazon's $97.9B Anthropic Convertible Notes May Make It IPO's Biggest Winner

Tema · AnthropicEstructural
2026-10-28 09:15 GMT+8

According to TechCrunch, Anthropic's IPO filing (S-1) indicates that tech giant Amazon is likely to be the single largest financial beneficiary of the listing. While traditional venture capital firms like Menlo Ventures and Sequoia are expected to return billions, none hold more than 5% of the company, excluding them from the principal stockholders' table. This is highly unusual for a tech IPO, where lead investors typically hold double-digit stakes.

Amazon's advantage stems from its unique investment structure. Since committing $4 billion in 2023, Amazon has structured much of its investment as convertible notes, debt instruments that convert into shares upon an IPO or sale. As of June 30, Amazon held $6.2 billion in principal outstanding. With Anthropic's valuation soaring to $965 billion, the shares these notes will convert into have become far more valuable. Amazon's own filings valued its note holdings at $97.9 billion.

Beyond the notes, Amazon invested $10 billion in nonvoting preferred stock in the second quarter, bringing its total investment to approximately $18 billion. Additionally, Amazon continues to earn significant revenue from Anthropic, which has committed to spending at least $110 billion on AWS over the next decade. Google also benefits similarly, holding an estimated 5.6% stake and securing $114 billion in cloud commitments.

This structure highlights the deepening capital and business ties between frontier AI labs and cloud providers. The S-1 warns that these cloud providers also build rival AI models and may not always align with Anthropic's interests. For investors, this signals a shift in wealth distribution in the AI sector from traditional VCs to strategic partners with infrastructure and capital power.

Fuentes:TechCrunch

Investigación

OpenAI Forces Cartoon Style for Election Interference Images in ChatGPT to Counter Deepfakes

Tema · OpenAIMaterial
2026-10-27 04:43 GMT+8

OpenAI has implemented a new image generation safeguard in its chatbot ChatGPT to address deepfake risks ahead of the upcoming US midterm elections. According to an October 20, 2026 update from OpenAI, if users request images "depicting election interference or that may discourage voting," the system will not refuse the request outright (provided it does not violate other content guidelines). However, it will force the output into an illustrated or cartoon style, ignoring all requests for photorealism.

The core logic of this mechanism is to lower the "evidentiary value" of such images. Josh Lawson, OpenAI's head of elections, told CNET: "The problem with deepfakes isn't social commentary, it's deception. So our aim is to degrade the evidentiary value of an image, while preserving room for illustrations and social commentary." By adding this visual friction, OpenAI aims to make the platform less useful for bad actors creating realistic political disinformation.

CNET reporter Katelyn Chedraoui tested the feature, finding that prompts involving sensitive scenarios, such as Immigration and Customs Enforcement (ICE) officers standing outside polling locations, consistently generated cartoon-style images. In contrast, Google Gemini only broadly stated it was rolling out additional protections without disclosing technical details, while Anthropic declared that using Claude to deceive voters violates its policies but did not offer a similar automatic style conversion mechanism.

As AI image models like ChatGPT Images 2.5 advance, the realism of generated content increases, making it harder for the naked eye to identify AI-created media. Existing verification or AI labeling methods are not yet industry-wide nor foolproof. OpenAI's strategy offers a practical alternative: sacrificing photorealism in specific political contexts to protect public trust. It remains unclear whether this measure will persist after the election cycle concludes.

Fuentes:CNET

Investigación

Xi'an Trade Surges 115.5% in First Three Quarters, Driven by AI Chip Exports

Tema · 西安AI芯片出口Material
2026-10-28 09:00 GMT+8

According to a South China Morning Post report, Xi'an, the capital of China's Shaanxi province, recorded a 115.5 per cent year-on-year surge in total trade during the first nine months of 2026, with exports climbing 142.1 per cent. This performance made it the top performer among provincial capitals and provincial-level cities for which data is publicly available.

Two product categories drove much of this export growth: shipments of automatic data processing equipment and parts jumped 388.1 per cent, while integrated circuits rose 251 per cent. Together, these categories accounted for 76.5 per cent of total exports. Zeng Gang, an economics professor at East China Normal University in Shanghai, noted that the surge reflects booming demand for chips used in artificial intelligence computing, alongside Xi'an's decade-long investment in the memory chip sector.

Operations by Samsung Electronics and Micron Technology in Xi'an stand out as immediate drivers. Samsung's Xi'an facility is its only overseas memory-chip fabrication plant, producing about 40 per cent of the group's NAND flash memory chips. The company invested 465.4 billion South Korean won (US$351 million) in the facility last year, a 67.5 per cent increase from the previous year. Micron announced plans in 2023 to invest more than 4.3 billion yuan in Xi'an over multiple years for acquisitions and new packaging-and-testing facilities.

Despite United States-led export restrictions on certain chipmaking equipment destined for China, both companies have continued investing in their Chinese operations. Official data showed that foreign-invested enterprises in Xi'an recorded 588.67 billion yuan in imports and exports for the first nine months, an increase of 174 per cent, representing 77 per cent of the city's total trade volume. The China-Europe Railway Express (Xi'an) was also cited as an indispensable pillar behind the rise.

However, the trade boom has not translated into robust overall economic activity. Xi'an's gross domestic product grew by just 1.1 per cent year on year in the first half of the year, 3.6 percentage points below the national average. Total retail sales of consumer goods declined by 3.6 per cent in the January-June period. Analysts warned that this growth model relies heavily on a small group of foreign semiconductor giants and is tied to the global semiconductor cycle, creating a divergence between export surges and sluggish local economic activity.

Fuentes:South China Morning Post

Investigación

Meta Raises AI Financing to $97.5 Billion; Nvidia and AMD Confirm Equity Stakes

Tema · Meta AI融资战略Material
2026-10-28 07:54 GMT+8

US social media giant Meta announced on October 27 that it has expanded the scale of its funding drive for artificial intelligence infrastructure. The company decided to increase the size of its previously planned Class A common stock public offering from $22.5 billion to $25 billion. This adjustment raises the overall total of Meta's current financing effort from the initially announced $95 billion to $97.5 billion (approximately 129 trillion Korean won).

According to the final pricing, the offering price for Class A common shares was set at $625 per share, a discount of approximately 2.7% from the closing price of $642 on October 23, corresponding to an issuance volume of 40 million shares. Concurrently, the accompanying convertible preferred stock offering was finalized at $50 per share for 250 million shares.

Market response indicates strong demand. According to Bloomberg, subscription orders for this additional issuance reached approximately five times the volume offered. More than half of the shares were allocated to long-term investors and existing shareholders, with the top 15 institutional investors absorbing 70% of the total. The convertible preferred stock also saw several-fold oversubscription.

Key partners' investment intentions have translated into actual commitments. Nvidia confirmed it will invest $3 billion in Meta's current stock offering, while Advanced Micro Devices (AMD) confirmed a $1 billion investment. Nvidia CEO Jensen Huang stated that the move aims to "optimize compute infrastructure from end to end" and expand the partnership with Meta. This marks a further deepening of capital ties between AI chip suppliers and platform providers.

Meta had previously announced on October 26 a comprehensive financing plan including common stock, convertible preferred stock, and newly created non-voting Class C shares to support its AI infrastructure expansion and development of Personal Superintelligence. Analysis suggests that Meta's third-quarter free cash flow turning negative (a deficit of $2.12 billion) prompted the company to seek larger external funding to alleviate financial pressure.

Fuentes:미래를 보는 창 - 전자신문

Investigación

HK Graduate Job Postings Drop 22% as 73% of Firms Plan to Require AI Skills

Tema · 香港应届生AI招聘趋势Material
2026-10-28 08:30 GMT+8

According to a report by the South China Morning Post, data from Jobsdb by SEEK, one of Hong Kong's largest job-search platforms, indicates that fresh-graduate job postings in the city fell 22 percent in the first nine months of 2026 compared to the same period in 2025. Concurrently, competition intensified, with a typical graduate listing attracting 56 percent more applications than a year earlier.

Bill Lee Chern-hsing, Managing Director of Jobsdb by SEEK in Hong Kong, noted that while finding a job has become more challenging, opportunities remain, albeit with higher requirements. A survey conducted by the platform in August and September involving 298 companies found that approximately 85 percent of respondents were willing to hire fresh graduates next year.

The survey revealed a structural shift in employer preferences. While 56 percent of employers focused on soft skills, followed by interview performance and work experience, a significant majority are adapting to technological changes. Specifically, 73 percent of companies plan to include artificial intelligence (AI) skills in their recruitment requirements for the coming year, with 54 percent of those firms stating they will place significantly more emphasis on AI capabilities than they did this year.

Alexa Chow Yee-ping, Managing Director of human resources consultancy ACTS Consulting, attributed the difficulty graduates face to the growing use of AI. She warned that weak demand is a global trend likely to persist or worsen as technology advances. She advised students to acquire AI skills through university courses or certifications and to consider fields less vulnerable to automation, such as sales, psychology, and counseling, which require a human touch.

Despite the shock to entry-level roles, the broader market outlook remains cautiously optimistic. Seventy-three percent of polled companies expect the 2027 employment market to be similar to or more active than in 2026, and 64 percent plan to maintain or increase their headcount.

Fuentes:South China Morning Post

Investigación

Linux Foundation Survey: 58% Cite Security Vulnerabilities as Top Data Infrastructure Risk; OpenSearch Production Usage Hits 36%

Tema · OpenSearchResumen rápido
2026-10-28 08:00 GMT+8

The Linux Foundation published its 2026 Open Data Infrastructure Survey results in October 2026. The study, based on responses from 294 participants across various industries and organizations, found that 58% of respondents identified "security vulnerability risks in data infrastructure" as their primary concern, surpassing regulatory compliance and data exposure risks (both at 33%). When evaluating data platforms, 79% considered security and compliance very important, a figure closely matching the 80% who prioritized Total Cost of Ownership (TCO), indicating that security considerations have become as critical as cost control.

The survey also tracked the adoption of OpenSearch, an open-source search and analytics platform managed by the OpenSearch Software Foundation under the Linux Foundation umbrella. Awareness of OpenSearch rose from 68% in 2024 to 89% in 2026. More significantly, the proportion of respondents using OpenSearch in production environments doubled from 19% in 2024 to 36% in 2026. Among the 128 current users, 83% utilize the platform for log analysis and observability tasks, while 52% employ it for security analytics, including threat detection and Security Information and Event Management (SIEM).

Fuentes:iThome

Investigación

Georgia Tech Researchers Develop ORISA Tool to Recover Deleted Local AI Chats from Android Memory

Tema · ORISAMaterial
2026-10-28 00:55 GMT+8

According to ZDNET, researchers at Georgia Tech in the United States have developed an experimental tool called ORISA that can recover deleted conversation records between users and local AI assistants from the running memory of Android phones. This finding has potential value for cybersecurity investigations but also reveals limitations in the privacy protection of so-called "local AI": even if chat sessions are deleted, some data may remain in the device's working memory.

ORISA works by leveraging batch tokens and key-value (KV) caches preserved in process memory to reconstruct token sequences and restore session-level attention states. In the experiment, researchers used the ChatterUI Android app to build a hypothetical note-taking AI assistant, inputted 100 pieces of fictional information about a project, and then deleted them. Subsequently, ORISA successfully recovered 26 usable pieces of information from memory remnants, enough to restart a new working session containing these facts—more than the 14 pieces the original assistant session could access due to falling outside its active memory.

Ken Xu, a Ph.D. student and lead author of the study, stated that this technology can help establish what information someone shared with an AI assistant and what assistance they received. The researchers noted that a downside of local AI is its lack of centralized logging and enforcement mechanisms, allowing bad actors to fine-tune models for crimes and then delete evidence to avoid tracking. ORISA aims to fill this cybersecurity forensic gap. However, this also means that if hackers acquire similar technology, they could theoretically gather deleted information from users' chats. Nevertheless, ORISA's efficacy depends on time, as it becomes harder to recover deleted information over time because the phone's memory gets overwritten by newer activity.

Fuentes:ZDNET

Investigación

Oracle Announces Business Semantic Layer and AI Data Catalog for Autonomous AI Lakehouse

Tema · 甲骨文自主运营AI数据湖Material
2026-10-27 14:31 GMT+8

At Oracle AI World 2026 held in Las Vegas on October 27, Oracle announced significant enhancements to its Autonomous AI Lakehouse platform. The core of these updates is the introduction of a 'Business Semantic Layer' and an 'AI Data Catalog', aimed at enabling AI agents to better understand the context and meaning of enterprise data.

The Business Semantic Layer allows organizations to define metrics, calculation methods, hierarchies, and Key Performance Indicators (KPIs) directly within the data platform. This enables AI agents to move beyond simple numerical retrieval to generating analytical results relevant to specific business tasks. By leveraging existing database annotations and AI-suggested analytical views, this layer adds business context to raw data, improving the accuracy of natural language queries and AI-driven analysis.

The new AI Data Catalog connects data stored in Oracle databases, third-party databases, external catalogs, and Apache Iceberg tables without requiring replication or movement. This provides a unified view of enterprise data assets. Both the Business Semantic Layer and the AI Data Catalog are expected to be available within the next 12 months.

Oracle also expanded support for Apache Iceberg, allowing users to create, add, modify, and delete Iceberg tables using SQL. This includes consistent analytical models across Oracle and Iceberg data, along with change history management and recovery capabilities.

In terms of performance, Oracle reported that its 'Data Studio' now supports connecting over 100 data sources for analysis using natural language, SQL, and Python. Additionally, the 'Data Lake Accelerator (DLA)' achieved up to 3 times faster query execution in internal benchmarks involving 100 terabytes of data. These performance figures are self-reported by Oracle.

Chet Ozvutun, Vice President of Autonomous AI Database Technology at Oracle, stated that simply providing AI access to corporate data is insufficient. He emphasized that for AI agents to make informed decisions and perform reliable tasks, they must understand the business context embedded in the data and operate according to organizational rules and policies. The new updates combine business semantics and multidimensional modeling with open data access and integrated governance to provide this necessary context and control.

Fuentes:zdnet.co.kr

Investigación

SK Hynix Q3 Revenue Up 328% YoY as AI Memory Demand Drives Record Profit

Estructural
2026-10-28 07:34 GMT+8

South Korean chipmaker SK Hynix released its third-quarter earnings report on October 27, 2026, showing that both revenue and operating profit sharply beat analyst estimates. Third-quarter revenue reached 104.62 trillion won ($78.86 billion), up 327.9% year-over-year, while operating profit surged 632.1% to 83.33 trillion won ($62.76 billion). These figures significantly exceeded LSEG SmartEstimates of 98.34 trillion won for revenue and 76.72 trillion won for operating profit.

As a key supplier of High Bandwidth Memory (HBM) chips to Nvidia, SK Hynix has emerged as one of the biggest beneficiaries of the global AI infrastructure buildout. Technology companies are ramping up spending on data centers and advanced processors, driving sustained demand for high-performance memory. Compared with the previous quarter, revenue climbed 31.9%, and operating profit surged 37.6%, indicating accelerating growth.

Despite the strong performance, some analyses noted that an increase in bit shipments offset a slowdown in average selling prices (ASPs), and the NAND business remained relatively weak. Additionally, SK Hynix announced last week that it would halt the review of potential IPO options for its U.S. subsidiary Solidigm, citing market volatility and share price trends. In August, the company also announced plans to repurchase and cancel 40 trillion won worth of its own shares, raising its shareholder return target to more than 50% of cumulative free cash flow generated between 2025 and 2027.

Fuentes:CNBC

Investigación

AI Security Firm Infisical Raises $40M Series B Led by Elad Gil at $600M Valuation

Tema · Infisical融资Resumen rápido
2026-10-27 08:00 GMT+8

AI security infrastructure company Infisical announced on October 27, 2026, that it has raised a $40 million Series B round at a $600 million valuation. The round was led by prominent Silicon Valley investor Elad Gil, with participation from Y Combinator, Gradient Ventures, Gallery Ventures, Pace Capital, and Dynamic Capital.

Infisical helps companies use AI agents without exposing their private credentials. As AI agents increasingly access production systems and perform actions previously requiring human approval, traditional security architectures struggle to keep up. Infisical's Agent Vault allows AI agents to access sensitive systems without revealing underlying credentials.

CEO Vlad Matsiiako stated in the official blog post that although the company has been cash-flow positive, they decided to raise funds to accelerate the building of next-generation software infrastructure, particularly focusing on security controls for AI agents. The capital will be used to expand the team (hiring over 100 positions) and deepen core offerings in secrets management, certificate management, and privileged access management.

Fuentes:Infisical

Investigación
Siguiente →
Iniciar sesión / Registrarse
LecturaInvestigaciónRadarMarco de inversión