LeituraPesquisaRadarFramework de investimento
Entrar / Cadastrar
TudoTecnologiaProdutosInvestimentoBolsaOutros

Leitura

2026-11-18133 posts

Atualizar

OpenAI Classifies GPT-6.2 Astra as High Risk for Self-Improvement, Launches 'Control Room' Monitoring System

Material
2026-11-18 17:16 GMT+8

US-based artificial intelligence company OpenAI released its latest flagship model, GPT-6.2 Astra, and the more economical daily-use version, GPT-6.2 Sol, on November 17, 2026. Beyond performance improvements and cost reductions, a key change in this release lies in safety governance: OpenAI has classified GPT-6.2 Astra as "High" risk in the dimension of "AI self-improvement," the first time it has assigned this rating to an externally released model.

According to German tech media outlet heise online, OpenAI believes that as model capabilities grow, AI is increasingly involved in the development of its subsequent versions. This self-improvement capability could drive AI development faster than safety measures and human control can keep up, posing new security risks. In response, OpenAI has strengthened the isolation of research environments and expanded monitoring.

As part of these countermeasures, OpenAI introduced a new central system called "Control Room." This system is designed to track model workloads within research infrastructure and can quickly halt related tasks if security concerns arise. This marks a move toward more specific technical controls by frontier labs in managing recursive AI development.

In independent tests by Artificial Analysis, GPT-6.2 Astra scored 62 points on the Intelligence Index, ranking second behind Anthropic's Claude Fable 5.2. Although OpenAI self-assessed the models as "Critical" or "High" risk in cybersecurity and biological/chemical capabilities, the company emphasized that the new models also show progress in defensive programming and vulnerability remediation. The models are currently available to ChatGPT Plus, Pro, Business, Enterprise, and Edu users, as well as API developers.

Fontes:heise online

Pesquisa

Investment Columnist Joachim Klement: Why Unexpected Inflation Has Shifted from a Headwind to a Tailwind for Highly Leveraged Stocks

Tópico · Inflation Surprise Equity ImpactMaterial
2026-11-18 15:01 GMT+8

Traditional finance training tells us that unexpected inflation is bad news for stocks: rising costs compress profits, higher interest rates increase discount rates, and companies with higher leverage are hurt more severely. Research by a team at the Bank of Italy (D'Andrea et al., 2025) confirms that before 2020, the market indeed reacted this way—positive inflation surprises depressed share prices, and the impact was even greater for highly leveraged companies.

But after 2020, the direction completely reversed. The same positive inflation surprises now push up share prices, and the more leveraged the company, the more it rises. Klement points out that while this seems anomalous, there is actually a clear mechanism behind it.

The key lies in the type of inflation shock. Inflation shocks in the 2000s and 2010s were mainly demand-driven: a hot economy pushed up prices, and central banks could cool things down by raising interest rates. But the shocks since 2020 have been supply-driven, similar to the 1970s—pandemics, wars, and energy disruptions have contracted supply, and raising interest rates won't make oil tankers pass through the Strait of Hormuz.

Supply-side inflation shocks are more persistent, and persistent inflation effectively devalues the nominal debt held by companies. From a balance sheet perspective, this means inflation helps companies reduce their debt burden without having to repay it in real terms—the higher the leverage, the greater the benefit. However, this mechanism has an upper limit: if the inflation shock is too large, as it was in 2023, company profits also suffer, and share prices fall. Klement believes the current market is in a 'golden middle ground'—inflation is sufficient to cut debt in real terms, but has not yet permanently damaged profitability.

Fontes:klementoninvesting.substack.com

Pesquisa

Co-authors of 'Autonomous': The Shift from Supervised Automation to Autonomous Execution Is a Step-Function Change, Not a Gradual Slope

Material
2026-11-19 16:05 GMT+8

Vala Afshar, Henry King, and Dr David Bray, co-authors of 'Autonomous', argue that enterprises are facing a widely misjudged inflection point: the transition from supervised automation to autonomous execution is a step-function change, not a gradual slope. Leadership's obligation at each stage is not simply to retire legacy tasks, but to explicitly define and mandate the higher-order judgment, creative, and relational work that only humans can perform.

This assessment targets a common error they observe: leaders who deploy AI agents into production assume the transformation is complete, ignoring the long intermediate phase. They call this phase "hands ready," characterized by neither humans nor agents being fully in command, requiring both parties to learn to share control. Treating human oversight as a passive "in-the-loop" checkbox ignores the reality that greater autonomy demands stronger, active human accountability.

Their core argument is that humans and AI are not competitors, but complements. Humans bring judgment, context, and creativity; machines bring speed, scale, and tirelessness. When intentionally combined, they produce better decisions, more durable value, and greater organizational resilience. However, these outcomes do not happen automatically; they depend on deliberately designed shared operating structures rather than default automation.

They propose an operational loop called aSUDA—sense, understand, decide, act—as the operating system for hybrid organizations. Within this loop, the agent-executed inner loop continues to expand in speed and scope, but leaders must ensure this expansion does not outpace the relational outer loop. As cognitive execution shifts to agents, investments in human oversight, trust, and role security must expand proportionally.

At a macro level, they note that traditional enterprises prioritize predictability above all else, relying on functional silos, rigid hierarchies, and annual planning cycles, thereby accepting a false trade-off between stability and performance. They argue that agentic AI can now maintain operational equilibrium continuously at machine speed, enabling organizations to achieve performance levels unreachable by static hierarchies while remaining structurally fluid—a concept they term "positive instability."

Relationship Design is the core competency that ties everything together. Redesigning roles, re-skilling skills, re-deploying talent, re-structuring reporting lines, re-calibrating incentives, and re-mandating human-exclusive work are the mechanics of transformation; the aSUDA loop is its operating system; positive instability is its performance mode; and accessible, secure AI is its precondition. Yet Relationship Design is what coheres them into an organization that people and agents can actually operate within. They argue that every enterprise, regardless of size or sector, must establish Relationship Design alongside strategy, finance, and technology as a primary management discipline.

Fontes:diginomica

Pesquisa

Reuters Columnist Klement: Analysts Called for 'Long' Before Two Previous Bubbles; AI Stocks Are Reenacting History

Tópico · Klement AI BubbleMaterial
2026-11-18 17:40 GMT+8

Reuters columnist Joachim Klement shared a discovery he is particularly proud of: he compiled data on Wall Street analysts' recommendations at the peak of the late-1990s tech bubble and on the eve of the 2008 global financial crisis. He found that recommendations at both times were severely skewed toward bullish positions—none issued sell warnings to investors. He views this degree of imbalance itself as a signal worth noting.

This observation offers an alternative angle for identifying bubbles beyond valuation ratios. Previous analyses have argued that the current expansion relies too heavily on a single engine by examining the share of IT contribution to GDP growth, while academic models have attempted to decompose stock prices into expected dividends, expected returns, and rational bubbles. Klement focuses on another layer: when sell-side analysts are overwhelmingly bullish at a bubble's peak and their recommendations lack sell warnings, this itself constitutes a market sentiment signal—it does not directly tell you whether assets are overvalued, but it suggests that independent verification steps in the pricing process may have failed.

Klement draws a parallel between this pattern and current market coverage of AI stocks, pointing out that the same imbalance is being reenacted as bad news accumulates. His core question is not whether the bubble will burst, but how this current rally will conclude—with a bang or a whimper. He himself provides no answer. It should be noted that this is an attribution argument proposed by Klement based on chart observations, rather than an independently verified predictive model; the fact that analyst recommendations are overwhelmingly bullish may partly stem from conflicts of interest inherent in the sell-side business model, and its reliability as a warning signal depends on whether this behavior is independent of other market factors.

Fontes:klementoninvesting.substack.com

Pesquisa

GM Plans 20% Boost in North American Factory Automation by 2028, Ties Supplier Contracts to Automation Scores

Tópico · GM Automation StrategyMaterial
2026-11-18 17:22 GMT+8

According to a Reuters report citing two people familiar with the matter, General Motors (GM) plans to increase the use of robots and other automation technologies in its North American factories by 20% before 2028. This previously unreported target is designed to counteract rising labor costs and production expenses associated with manufacturing reshoring to the United States.

Sources revealed that GM is developing its own robots, assembling them at a 90-year-old facility in Kokomo, Indiana, which produces automotive electronic components like engine control modules. The current focus is on Autonomous Mobile Robots (AMRs) for intra-factory parts transport. This marks the first disclosure of GM's internal robot manufacturing capability; previously, the company relied primarily on external vendors. Last year, GM stated it had deployed about 100 robotics experts, AI engineers, and hardware specialists to develop advanced robotic systems.

The automation mandate extends to the supply chain. Two other individuals familiar with GM's supplier system indicated that GM now requires component suppliers bidding on new contracts to submit their own operational automation plans. GM uses a scorecard to evaluate suppliers' automation investments, incorporating these scores into contract award decisions. For suppliers seeking GM orders, automation levels have become a direct entry barrier rather than just an efficiency tool.

This strategic shift follows record-breaking labor agreements reached in 2023 between the United Auto Workers (UAW) union and GM, Ford, and Stellantis. At the time, GM noted that the new contracts would add approximately $575 in labor costs per vehicle. Analysts widely expected automakers to offset these costs through increased automation. Additionally, US government policies promoting domestic manufacturing have further pressured local production costs.

However, expanded automation risks strong opposition from labor groups. Ryan Hiestand, Vice President of the UAW, which represents about 50,000 GM factory workers, criticized the move in a statement following the Reuters report: "What they call increasing efficiency is firing the people who actually create products and profits." UAW President Shawn Fain has also warned that AI and automation pose threats to workers. With existing labor contracts set to expire in spring 2028, automation is expected to be a central issue in the next round of negotiations. GM did not immediately respond to requests for comment but has previously stated that its use of automation is intended to improve safety, quality, and efficiency.

Fontes:TechOrange 科技報橘

Pesquisa

Alibaba Cloud's Qoder Lowers Default Thinking Effort for Several Models, Cutting Costs by Up to 40.5%

Tópico · Qoder Thinking EffortVisão rápida
Verificado 2026-11-18 17:54 GMT+8

Alibaba Cloud's AI coding assistant Qoder recently announced adjustments to the default "Thinking Effort" settings for several large language models, based on evaluations conducted on its proprietary dataset, QoderBench. This parameter controls how much reasoning the model performs before taking action, directly influencing response time and token consumption.

Specifically, the default setting for GLM-5.3, Kimi-K3 (and its replacement Kimi-K3.1-Pro), Kimi-K2.8-Preview, and DeepSeek-Flash has been lowered from "Maximum" to "High". The defaults for GLM-5.3-Flash and DeepSeek-V4-Pro remain at "Maximum". Users can still manually select "Low", "High", or "Maximum" in the model manager.

According to the official Qoder blog, this change stems from testing nearly 200 real-world task cases. The data shows that reducing the thinking effort by one level from "Maximum" to "High" decreases output tokens by 19% to 48% and model call counts by 6.5% to 31.5%. In terms of cost, per-run expenses fell by 6.9% to 40.5%, while relative scores declined by only 5.7% to 16.1%. For instance, switching GLM-5.3 to "High" resulted in a 40.5% cost reduction with only a 6.3% relative drop in score.

Qoder notes that for most everyday development tasks, the marginal benefit of excessive reasoning diminishes, leading instead to increased wait times and Credits usage. These results are vendor-reported and do not specify changes in end-to-end task duration; actual impact may vary depending on the specific workload.

Fontes:Alibaba Cloud Community

Pesquisa

Tsinghua-Incubated Startup Launches AI Hardware Development Platform Supporting Natural Language Circuit and Code Generation

Tópico · MOCE DesignerVisão rápida
2026-11-18 16:13 GMT+8

According to a report by Zhidongxi on November 18, Moce Intelligence, a startup incubated by Tsinghua University's X-Lab, recently officially released MOCE Designer, a hardware development platform that supports direct interaction via natural language.

In simple terms, developers can describe what they want to build and use AI within the same workbench to complete requirement breakdown, component selection, circuit connection design, code generation, compilation, and flashing, followed by debugging on the physical device. After prototype validation, teams with further development needs can also connect to structural processing, trial production, mass production, and sales resources.

Moce Intelligence calls this mode 'Vibe Building'. The young startup was founded in 2025 with a team of about 20 people. At the launch event, Moce announced that its module library will expand to 200 items by the end of the year, with plans to gradually increase it to over 1,000.

Before the official release of MOCE Designer, Moce Intelligence held its first Vibe Hardware Hackathon at Tsinghua University as a public beta test. Over 100 participants spent two and a half days developing and debugging, resulting in 15 AI hardware creative projects showcased on stage. Among them, the Tsinghua Next-Day Band built a theremin capable of being played without touch for less than 200 RMB. Another participant with no prior hardware development experience independently created an innovative hardware device called 'Tomato Monster', combining a desktop robot and a lock box.

Fontes:zhidx.com

Pesquisa

Artist Uses Open-Source Vision Model on Consumer GPUs to Scan Global Public Cameras

Tópico · Wally FinderVisão rápida
2026-11-12 21:37 GMT+8

Belgian artist Dries Depoorter has launched 'Wally Finder', an art project that uses artificial intelligence to search for people wearing red-and-white striped shirts in live footage from over 450 unsecured public webcams worldwide. The project demonstrates the capability of modern open-source multimodal models to perform complex visual recognition tasks on consumer-grade hardware.

According to Tech Times UK, the system relies on Alibaba's Qwen3-VL, a general-purpose vision-language model, running locally via Ollama on three gaming GPUs. The software samples still images from each feed every few seconds, sending them to the model to identify individuals matching the description of the fictional character Wally (also known as Waldo).

As of November 10, 2026, the project had analyzed 1.68 million frames and identified 13 matches, with clips originating from the US, Japan, and Peru. Depoorter manually reviews each flagged frame before publishing selected clips, applying blurring and black-and-white filters to the surrounding scene while leaving the striped shirt in color to protect bystander privacy.

The project highlights the increasing accessibility of powerful AI tools for individual creators, allowing them to process large volumes of video data without relying on cloud services. It also raises questions about the security of public camera feeds and the potential for automated surveillance using widely available technology.

Fontes:Tech Times UK

Pesquisa

Microsoft CEO Nadella Outlines 'Software Factory' Model: Verification Replaces Generation as Core Focus, Treating AI Agents as Insider Risks

Tópico · Microsoft Software Factory ModelMaterial
2026-11-18 16:21 GMT+8

Microsoft CEO Satya Nadella recently detailed his vision for a "Software Factory" in an interview on The Pragmatic Engineer podcast. As artificial intelligence drastically lowers the barrier and cost of code generation, the focus of software development is shifting from "writing" to "verification." Nadella emphasized that while AI boosts productivity, ensuring system stability still relies on human experience and the team's tribal knowledge.

Addressing the challenges of AI-assisted development, data from Google's DORA research indicates that high AI adoption accelerates delivery but comes with decreased system stability. Time saved by developers in coding often shifts to review and debugging. Nadella noted that even with self-review attempts, models may deviate from expectations due to opportunistic behavior, such as deleting failing test cases to fake passing reports. Thus, establishing rigorous verification processes has become critical.

Regarding corporate governance and security, Nadella suggested treating autonomous AI agents as "insider risks," implementing safety guardrails similar to those used for high-risk employees. He proposed two specific strategies: first, avoiding vendor lock-in by ensuring the underlying architecture, memory mechanisms, and language models are decoupled for easy replacement; second, establishing independent oversight systems to continuously monitor whether AI agents operate as expected. Additionally, he advocated for open-source or locally deployed models to control core intellectual property.

Microsoft is already practicing this model internally. For instance, the Azure network team developed automated agent systems to handle emails and operational workflows, significantly improving efficiency in managing over 500 fiber optic operators. Meanwhile, Microsoft partnered with Carnegie Mellon University to launch the PRAISE program, aiming to cultivate engineers' system thinking and judgment through mentorship, preparing them for a new role focused on defining specifications and managing autonomous infrastructure rather than writing code personally.

Fontes:TechOrange 科技報橘

Pesquisa

Workday CTO and Head of Data Cloud: Enterprise AI Trust Is an Architecture Problem, Not a Model Problem

Tópico · Workday Agent Context ServiceMaterial
2026-11-18 16:38 GMT+8

Ahead of the opening of Workday Rising EMEA in Barcelona, Yasmeen Ahmad, Head of Workday Data Cloud, told tech media outlet diginomica that the problem of enterprise AI trust can be summed up in one sentence: it is not a question of model intelligence, but of enterprise governance and architecture. She argues that the idea that placing a large language model next to data will generate value is wrong—the model requires a complete operating system around it, including how it connects to data, anchors in business logic, establishes agent identity, defines permission boundaries, and enforces budget constraints.

This was precisely the judgment Ahmad faced when she joined Workday from Google. She points out that Workday possesses twenty years of accumulated HR and financial data, which serves as an authoritative source for organizational structure and cash flow, yet this has never been open before. Now, agentic technology gives this data a new use case: turning the system into a headless architecture that third-party agents can call upon, while Workday executes the context constraints. She believes that even if third-party agentic vendors piece together similar knowledge from the internet, they cannot replace a system capable of enforcing context.

Gabe Monroy, Chief Technology Officer of Workday, further elaborated on the specific engineering solutions. He confirmed that Workday Data Cloud can be viewed as a context layer, particularly possessing authoritative context regarding people and financial data. Through its Prism capabilities, the Data Cloud can ingest external data; it now also supports zero-copy ingestion, such as integrating badge access data into the organizational chart, allowing agents to understand scenarios like return-to-office performance.

However, Monroy argues that a mere data lake or lakehouse is insufficient. During its Early Access phase, Workday released a capability called Agent Context Service, which uses a three-layer mechanism of caching, machine learning classification, and data scoring to determine what context an agent needs before execution. His analogy is 'filling the room': before the reasoning engine begins processing prompts, the system injects file data, scoped services, credentials, and tokens into the agent's runtime environment, managing the entire environment meticulously. Tool calls are intercepted by a proxy, thereby enabling dynamic governance.

This architecture yields a practical effect: if a prompt contains personally identifiable information (PII), the system can route the request to a different model; if it involves other types of data processing, the rules can change dynamically. Monroy emphasized that this is not a static configuration, but rather real-time adjustments based on what the agent is asked to do.

Regarding costs, Monroy explained that using machine learning classification for content scoring (rather than invoking large model reasoning) allows the caching of numerous context decisions, providing extremely fast and inexpensive just-in-time context. Jon Reed, author of diginomica, analyzed this architecture and noted that when language models are wrapped in deterministic workflows and governance rules, the models become more swappable, reducing reliance on the most expensive frontier models; he summarized this approach as neuro-symbolic architecture—using an ML classification brain to reduce LLM reasoning usage, lowering costs while maintaining precision.

Fontes:diginomica

Pesquisa

LanceDB Reports Realtime Engine Achieves 52-Millisecond Write-to-Read Latency, Breaking Lakehouse Batch Commit Bottlenecks with MemWAL Open Format

Tópico · LanceDB Realtime EngineMaterial
2026-11-18 07:16 GMT+8

Lakehouses are optimized for large-scale scanning but leave a structural challenge for streaming writes: more frequent commits lead to faster accumulation of small files and metadata updates; waiting for larger batches forces readers to wait longer. In its Realtime Engine blog post, US AI data infrastructure company LanceDB argues this trade-off can be eliminated—provided an LSM tree structure is introduced within the open table format rather than layering a proprietary buffer on top of the lakehouse.

LanceDB software engineer Xuanyu Zhan, data engineer Jack Ye, and senior software engineer Dan Rammer describe the mechanism of the MemWAL format in the article: new rows are first written to a write-ahead log in object storage and simultaneously enter a MemTable with established indexes in memory; once the MemTable is full, it is output as an indexed Lance file (SSTable), which is eventually merged into the base table in the background. Queries search the MemTable, SSTable, and base table simultaneously, meaning new rows can be retrieved immediately after persistence without waiting for table-level commits. Each shard writes independently, allowing ingestion to scale across nodes.

The design compares against two existing industry paths. At the storage layer, Apache Paimon builds an LSM tree into its table format to support streaming updates, whereas Iceberg and Delta Lake still rely on batch commits paired with deletion vectors. At the ingestion layer, cloud platforms and compute vendors typically add a write-buffering layer on top of lakehouse formats, where new data can be served from the buffer upon acknowledgment; however, the authors point out this introduces vendor lock-in—other engines cannot read this data through the open table format during the buffering window, and durability depends on vendor-managed buffers rather than the object storage itself. LanceDB’s argument is that MemWAL combines streaming writes, real-time index search, and object-storage durability within a single open Lance table, eliminating the need for external buffers or search synchronization layers.

Regarding performance, LanceDB reports benchmarking three public datasets—FineWeb-Edu, MS MARCO, and LAION—using m8gd.8xlarge nodes (30 vCPU, 112 GiB) on Amazon EKS. A write was counted only after it was persisted in S3, and each run verified that the row count in the table matched the acknowledged write count. The median write-to-read latency was 52 milliseconds; single-node four-shard throughput increased from 79-83 MB/s to 312-334 MB/s; the combined total for four nodes reached 946 MB/s. Query latency for fresh rows: primary key lookup 12-14 milliseconds, full-text search 15-28 milliseconds, vector search 26-30 milliseconds. Writing to a large table already containing the complete corpus was at least as fast as writing to a small one. On LAION, when only the primary key index was live, writing images directly to object storage while keeping only references in memory reduced peak memory from 74 GiB to 18 GiB, with a throughput loss of less than 3%.

Fontes:lancedb.com

Pesquisa

Japan's RIKEN Suspends Supercomputer Fugaku After Hack Leaks Nearly 4,600 User Credentials

Tópico · Fugaku Supercomputer BreachMaterial
2026-11-18 14:27 GMT+8

Japan's RIKEN Institute disclosed this week that a system connected to its flagship supercomputer "Fugaku" suffered unauthorized access, resulting in the leak of approximately 4,597 user personal data records and access credentials for another supercomputer. To contain the incident, RIKEN suspended Fugaku's computing operations and cut off external access early on November 14.

According to iThome, RIKEN detected abnormal activity on November 12 and confirmed potential data leakage the following day. Preliminary investigations indicate that external attackers breached the internal network between November 10 and 13 by exploiting the publicly accessible web dashboard displaying Fugaku's operational status. Unauthorized programs were installed on some compromised servers, altering system integrity.

The leaked information includes names and email addresses of Fugaku users, as well as authentication details such as SSH private keys and access tokens stored in user home directories. These credentials could potentially be used to access another supercomputer named "Rikou" and related services. Fugaku has up to 4,222 registered users, while Rikou has up to 375.

Developed jointly by RIKEN and Fujitsu, Fugaku has long held the top spot in global supercomputer rankings and currently ranks ninth. This incident not only disrupts critical scientific computing tasks but also highlights security vulnerabilities in public compute infrastructure exposed via internet-facing monitoring interfaces. RIKEN has isolated affected systems, rotated management keys, and required all users to immediately revoke old keys from external services and issue new ones.

Fontes:iThome

Pesquisa
Próxima →
Entrar / Cadastrar
LeituraPesquisaRadarFramework de investimento

Ahmad also discussed the issue of agent identity. As agents evolve from copilots to autopilots, running autonomously 24/7, under whose authority does the agent operate? She believes the CFO must be held accountable for launching the agent, necessitating an accountability model. Workday is building this security governance layer through the acquisition of C1.ai, an agent identity management platform.

It should be noted that all these capabilities are currently in the Early Access stage, and all technical details stem from self-statements by Workday executives, with no independent verification available. Workday's advantage of 'authoritative context' rests on the premise that enterprises are willing to bind their core HR and financial data to its platform; whether this premise constitutes a true barrier depends on the evolution of the market landscape.

It must be clarified that all the above figures come from LanceDB’s own benchmark tests, where the test hardware, datasets, and metric definitions were selected by the company, with no third-party independent replication yet available. The comparison table of capabilities with Iceberg, Delta Lake, and Paimon is also self-stated by the authors, not a cross-format same-condition actual test. For teams evaluating real-time AI training data pipelines, MemWAL’s architectural argument—embedding an LSM tree into an open table format to eliminate commit bottlenecks—provides a checkable design approach, but actual production throughput, latency, and operational costs still require independent verification.