LectureRechercheRadarCadre d'investissement
Connexion / Inscription
ToutTechnologieProduitsInvestissementBourseAutres

Lecture

2026-11-1298 publications

Actualiser

Nvidia CEO Removal Rumors Trigger AI Supply Chain Jitters: Micron and SanDisk Stocks Fall Pre-Market

Matériel
2026-11-12 19:58 GMT+8

On Thursday, November 12, shares of US memory chip manufacturers Micron Technology and SanDisk fell in pre-market trading, partially reversing the previous day's strong rebound. The decline was triggered by media reports suggesting that Nvidia's board of directors was considering the removal of Chief Executive Officer Jensen Huang, unsettling investors regarding the stability of the broader artificial intelligence supply chain.

Data shows Micron traded at $790.92, down 2.51%, while SanDisk fell to $948.88, a drop of 2.95%. This followed significant gains of 11.89% and 13.69% respectively during Wednesday's regular session, driven partly by CoreWeave's latest results which had eased fears about slowing momentum in large AI infrastructure projects. However, sentiment shifted overnight as Nvidia's own stock plunged more than 8% after hours.

Despite short-term volatility, fundamental data offers some support. Micron's fourth-quarter revenue reached $54.23 billion, up from $11.32 billion a year earlier, reflecting strong demand for high-bandwidth memory (HBM) and rising prices. SanDisk's first-quarter fiscal revenue hit $11 billion, up 23% sequentially, with management forecasting $12.2 billion to $12.6 billion for the next quarter. Analysts note that if AI spending merely redistributes among suppliers rather than contracting, Micron, as a key HBM provider, may prove more resilient than SanDisk, which is more exposed to general storage market cycles.

Sources :invezz.com

Recherche

JD.com Completes Voluntary Public Takeover Offer of CECONOMY

Matériel
2026-11-12 20:01 GMT+8

JD.com announced on November 12, 2026, that it has completed its previously announced voluntary public takeover offer for CECONOMY AG. The transaction resulted in JD.com acquiring 59.8% of CECONOMY's share capital and voting rights for a total consideration of EUR 1.335 billion. As a result, CECONOMY becomes a subsidiary of JD.com, and its financial results will be consolidated into JD.com's consolidated financial statements starting from November 12, 2026.

Together with the 25.3% stake retained by JD.com's partner Convergenta, this acquisition brings the combined shareholding in CECONOMY to 85.2%. The offer was made through JINGDONG HOLDING GERMANY GMBH, an indirect wholly-owned subsidiary of JD.com, at a price of EUR 4.60 in cash per share. The completion follows the fulfillment of all offer conditions, including merger control clearances, foreign investment clearances, and EU foreign subsidies clearance.

CECONOMY is one of Europe's largest consumer electronics retailers, operating more than 1,000 stores across 11 markets under the MediaMarkt, Saturn, and MediaWorld brands. The business employs over 50,000 people and reported more than 53 million loyalty members and 2.2 billion annual customer contacts in 2025. Sandy Xu, CEO of JD.com, stated that the two companies share a vision to build Europe's leading next-generation consumer electronics platform, combining CECONOMY's local expertise with JD.com's technology and logistics capabilities.

As part of the strategic roadmap, CECONOMY will remain a stand-alone business in Europe with a local independent technology stack. No changes are planned to the workforce, employee agreements, or sites, aiming to preserve the brand heritage and local advantages while leveraging JD.com's infrastructure for efficiency improvements.

Sources :GlobeNewswire News Room

Recherche

Susannah Streeter, Chief Investment Strategist at Wealth Club: Huang’s Scandal Exposes Key Person Risk for Nvidia, Valuation Multiples Face Compression

Matériel
2026-11-12 19:35 GMT+8

Nvidia's stock price has recently come under pressure, but Susannah Streeter, Chief Investment Strategist at Wealth Club, believes that what truly warrants investor attention is not the short-term decline, but rather the pricing logic of the valuation premium being re-examined. Speaking on the podcast Zero Sum, she pointed out that Nvidia's position as a core supplier of AI infrastructure means a significant portion of its valuation premium is bound to CEO Jensen Huang personally—he helped investors understand the scale of the AI opportunity and positioned Nvidia as the central computing infrastructure supporting this wave of construction. When questions arise regarding Huang's judgment, this binding relationship creates a disproportionate impact.

Streeter's core mechanism is that even if chip demand remains robust, a shock to leadership confidence could trigger compression of valuation multiples. Investors will require the company to demonstrate a broader leadership bench and clearer governance structures, while regulators and lawmakers are tightening scrutiny around AI and chip exports. This means the premium is no longer supported solely by revenue growth and market share; it also requires governance credibility as a prerequisite. She cited Apple's transition from Steve Jobs to Tim Cook as an example, illustrating that founder-driven companies can enter new phases, but they need different types of leaders to guide the company to the next stage.

The second-order effects of this incident extend to US policy on chip exports to China. Streeter argues that the scandal has added a national security dimension to an already highly politicized debate. She expects lawmakers to seek greater oversight of advanced chip exports, including measures requiring companies to track the overseas flow of chips to prevent smuggling. Washington may also pressure allies to restrict exports of semiconductor manufacturing equipment to China. With Democrats controlling both the House of Representatives and the Senate, she anticipates increased pressure for AI regulation and more restrictions on the influence of large technology companies.

For investors, Streeter's advice is not to immediately sell off Nvidia shares, but to reassess portfolio concentration. She warns that investors may have exposure to a select group of US technology companies through global funds that exceeds their expectations, and they should ensure adequate diversification across industries and geographies. She simultaneously distinguishes between short-term volatility and the long-term prospects of the AI industry: Nvidia remains a key supplier with strong demand, but not every company associated with the AI boom will become a long-term winner—she draws an analogy to the dot-com bubble period, where many star companies failed while a few, such as Amazon, became true beneficiaries.

The applicability boundaries of Streeter's argument lie in the speed of the transition and the degree of regulatory escalation. Following her analytical logic, if Nvidia's board quickly completes a credible leadership transition and export policies do not tighten further, the company's profitability and premium may be re-established. Furthermore, she maintains a cautious stance on infrastructure investment—companies investing huge sums without proving sufficient returns face valuation pressure, whereas underlying infrastructure suppliers such as power, grid, and utilities will still receive support even if data center construction slows. Her overall judgment is that the investment narrative for the AI sector is shifting from simple demand growth to a comprehensive assessment of governance quality, regulatory environment, and capital return rates.

Sources :invezz.com

Recherche

Monks Investment Trust Quarterly Review: Winners in the AI Supply Chain Depend on Where the Next Bottleneck Is

Matériel
2026-10-18

There is a time-dimension issue that is easily overlooked in AI infrastructure investment. When a specific link becomes the most severe bottleneck, companies that solve it earn excess returns—but these high returns themselves attract capital to fill that bottleneck. Once the bottleneck is resolved, excess profits compress. The real question is not who benefits from current AI demand, but where the next constraint will appear.

In its October 2026 quarterly review, Monks Investment Trust explicitly drew this distinction. It wrote that so far, the biggest winners have been suppliers addressing the most severe bottlenecks, but many chokepoints may be relatively easy to fix, and the exceptional returns they generate may not persist. The fund’s core AI holdings—NVIDIA, Taiwan Semiconductor Manufacturing Company (TSMC), and Amazon—are positioned as durable building blocks of the system, rather than merely beneficiaries of today’s shortage.

Around this core, Monks tilted toward companies solving capacity constraints this quarter. Specifically, it added to positions in four semiconductor equipment companies, including Disco (whose equipment grinds and polishes wafers) and Kokusai (whose machines deposit layers used in memory chips), citing their dominance at critical steps in the manufacturing process. As technology evolves, new constraints are emerging: Astera Labs’ connectivity chips move data between expensive AI chips to ensure they do not sit idle; SoftBank provides undervalued exposure to the logic chips needed for agent coordination through its holding of Arm, a logic chip designer.

The technical basis for this framework is that AI agents have already surpassed human usage and are expanding rapidly, despite Cisco President Jeetu Patel noting that less than 2% of humanity uses agents today. As agents become a larger share of AI usage, logic chips capable of coordinating and directing their actions become increasingly important. This is not an abstract investment thesis, but an observation of the evolution of AI system architecture—from computation to connectivity, from connectivity to coordination, with each layer solved exposing the constraint of the next.

It must be stated honestly that this review was written against the backdrop of Monks underperforming the benchmark over the past 12 months (share price return 11.9%, net asset value return 10.3%, compared to the FTSE World Index’s 20.9%). The authors explicitly state that relative underperformance is partly due to different AI exposures. The argument about the 'next bottleneck' is difficult to falsify in the short term—if these companies indeed become winners over the next five years, the framework is validated; if not, it can also be reinterpreted as the 'next-next bottleneck'. The authors have a clear commercial interest in providing a narrative justification for underperformance.

But this analytical framework itself holds value independent of the fund’s performance. It poses a testable question: in the capital expenditure cycle of AI infrastructure, which constraints are temporary (high returns attracting capital to quickly fill them) and which are structural (requiring sustained technological leadership)? Answering this question requires understanding specific manufacturing processes, chip architectures, and system topologies, rather than simply tracking total AI spending.

Sources :bailliegifford.com

Recherche

Avramov et al. Find: Highest-Alpha Factor Strategies Have Capacity Under $2 Billion, Profitability Factors Support $33 Billion

Sujet · 因子策略容量之争Matériel
2026-11-12 17:33 GMT+8

Quantitative investment literature has long evaluated whether a signal is worth pursuing using metrics such as alpha, the Sharpe ratio, and statistical significance. A new paper by Avramov, Bongaerts, Crego, and Kvaerner introduces an often-overlooked dimension: capacity—the maximum amount of capital a strategy can deploy before transaction costs erode its market-hedged returns. Using NYSE TAQ data from 2010 to 2023, they estimated spreads, price impact, and liquidity replenishment to calculate deployable capital for 35 characteristic-based anomalies and factors one by one.

The core finding is a severe disconnect between alpha rankings and capacity rankings. The two long legs with the highest gross alphas each support less than $2 billion in market-hedged capital, while strategies with significantly lower alpha in profitability and cash flow can support $20 billion to $33 billion. The profit-margin anomaly has roughly two-fifths the alpha of the highest five-day return strategy, yet its capacity is 97 times greater. For anyone actually allocating capital, this means that the most eye-catching signals in backtests are not necessarily the largest economic opportunities.

The differences in capacity stem primarily from price impact, which is driven by two portfolio characteristics: turnover and the liquidity of the underlying stocks. Short-term reversal strategies turn over 1.8 times per month, incurring 1.6% monthly impact costs at a $10 billion scale; the Amihud liquidity strategy hardly turns over (0.12 times per month), but because it holds illiquid stocks, it still incurs 1.2% monthly impact costs at the same scale. The paths differ, but the result is the same: costs eat away at alpha.

The paper also provides a lever with significant practical implications: rebalancing speed. When the adjustment coefficient τ is less than 1, exiting stocks that leave the target quintile does not involve immediate liquidation but rather a gradual reduction over subsequent months. This single change raises the median capacity of positive-capacity factors from $1.5 billion to $21 billion, a thirteenfold increase at the strategy level. The value-weighted machine-learning composite strategy expands its capacity from $177 million to $19 billion without any change to the signal itself. The rank correlation of capacity rankings is 0.93, indicating that slower rebalancing mainly amplifies scale without drastically altering relative ordering.

It is crucial to emphasize that these figures come entirely from the authors' own first-party estimates, based on a specific data window and specific assumptions about market hedging costs, and have not yet been independently replicated. The median factor capacity of $1.53 billion is in the same order of magnitude as the median hedge fund's net asset value of $1.1 billion; this comparison itself serves as a sobering reminder of scale constraints: most factor strategies are not frictionless at institutional sizes.

The practical implication for readers is: before deciding how much to allocate to an anomaly or factor, measure its trading capacity, rather than looking only at backtested returns. The gap between a niche $200 million strategy and a $20 billion institutional strategy is written in the liquidity of the stocks traded and in the speed of rebalancing.

Sources :quantpedia.com

Recherche

"AI-Driven Data Foundation Design" Author Sanjeev Mohan: Agents Bypass Pipelines to Build Their Own Databases, Observability Must Be Embedded in Architecture

Sujet · AI智能体数据访问模式Matériel
2026-11-12 19:02 GMT+8

For decades, enterprise data architecture followed a fixed path: data was generated once in business systems, then pushed through pipeline stages such as extraction, transformation, cataloging, and preparation before being consumed by humans for batch reports. Sanjeev Mohan believes the premise of this model is now being broken. His assessment is that when thousands of agents run simultaneously, they will not wait for pipelines to finish—they demand direct access to the database.

Mohan describes a behavioral pattern never before seen in software: agents create short-lived database instances, copy data within them, parallelize tests across different reasoning chains, and only then select a course of action. He cited a loan approval example: an agent might query census records and social media data simultaneously before rejecting an application. This immediately raises a governance question—how is that decision audited?

In his view, this differs fundamentally from past data governance. Past governance was deterministic: manage the data, then run a fixed reporting dashboard on top of it. Agent governance faces constantly changing context at runtime. For the same task, the context could be entirely different depending on the person (or the specific agent instance). Therefore, he stated that observability and transparency must become intrinsic parts of the architecture, rather than things bolted onto a data warehouse afterward.

From this, Mohan derives a further conclusion: future platforms will need to fuse structured and unstructured data into a single converged system, and new types of databases will emerge—specifically designed to store runtime state and build runtime context. He has not yet provided the specific form or implementation path for such databases; this judgment remains a proposal regarding architectural direction.

It should be noted that all of the above are personal arguments articulated by Mohan during an interview with theCUBE at KubeCon North America. As the head of SanjMo Consulting and author of "Designing the AI-Driven Data Foundations," he has long studied data architecture issues. However, the text provides no measured data from deployed systems or independent verification. The value of this argument lies in its precise naming of a constraint: agents' data consumption patterns differ from those of human consumers, meaning pipeline architectures designed around human consumption will constitute bottlenecks.

Sources :SiliconANGLE

Recherche

VulnCheck Report: Only 217 of Nearly 40,000 Candidate Vulnerabilities Discovered by Anthropic's Glasswing Are Publicly Fixed

Matériel
2026-11-12 15:55 GMT+8

Cybersecurity firm VulnCheck released a report in November 2026 independently tracking Anthropic's "Project Glasswing," an initiative designed to use Claude models to discover security vulnerabilities in open-source software at scale. The report indicates that seven months after the program's launch, Claude had identified 37,966 candidate vulnerabilities, yet only 217 were marked as fixed in public records as of early November.

This data contrasts with figures disclosed by Anthropic itself. According to official data from November 2, out of 37,966 candidate findings, 6,376 had been reviewed by external experts, 5,087 were reported to maintainers, and 629 received upstream patches. While differing statistical methodologies account for some discrepancies in the number of fixes, both parties acknowledge a core issue: the speed of AI-driven vulnerability discovery has created a severe bottleneck in human review capacity, significantly limiting disclosure and remediation efficiency.

To alleviate this pressure, Anthropic launched the OSS Scanner service on October 8, allowing open-source maintainers to receive model-generated vulnerability reports directly, bypassing Anthropic's internal human review process. However, this "zero human intervention" model shifts the validation burden to the community. For instance, Daniel Stenberg, maintainer of the curl tool, recently received 45 vulnerability reports from Anthropic. He stated that the core maintenance team was overwhelmed by the influx of reports and had to change rules to discuss low and medium severity vulnerabilities publicly on GitHub, leveraging community help for review.

VulnCheck's report notes that this gap is accelerating. In just the last two months, candidate vulnerabilities surged by nearly 12,000, while the record of fixed issues increased by only 15. Additionally, the 442 CVEs attributed to Anthropic tracked by VulnCheck represent only 0.5% of the global public CVE total for 2026. Although Anthropic did identify serious flaws like a Use-After-Free memory error in curl via Claude, the overall picture suggests that AI's powerful discovery capabilities have not translated into proportional remediation outcomes, instead imposing a heavy review workload on open-source maintainers.

Sources :量子位

Recherche

Liao Hung-yuan, Director of the Institute of Information Science, Academia Sinica: Large Models Can Describe Images but Fail to Understand Human Actions; Modular Visual Understanding Is Needed to Bridge the Gap

Sujet · 廖弘源模块化视觉理解Matériel
2026-11-12 18:04 GMT+8

When a photo of a child is input into a large vision-language model, the model’s description is merely "Everything is normal." Liao Hung-yuan, Director of the Institute of Information Science at Taiwan's Academia Sinica, uses this example to illustrate that while current general-purpose vision models possess considerable image-description capabilities, they struggle with more nuanced judgments closer to real-world needs—such as what the person is doing, whether danger is occurring, or what relationships exist between objects in the environment.

Liao Hung-yuan believes that to bridge this gap, Academia Sinica’s direction is not to build another large general-purpose model, but to develop specialized visual-understanding capabilities that complement large language models, including behavior analysis, object understanding, and edge computing. His reasoning is straightforward: edge devices have limited computing power and memory, making it impossible to run large models directly on cameras. Therefore, multimodal AI must adopt a modular approach, allowing different models to handle tasks suited to their strengths.

Home care is the most intuitive validation scenario for this architecture. When a camera detects someone falling, the system cannot simply know "there is a person in the frame"; it must determine if the person has fallen and requires assistance. If help is deemed necessary, the system uses text-to-speech to ask, "Do you need help?" After the user responds, automatic speech recognition interprets the voice content, which is then connected to a large language model. Finally, the system notifies family members or emergency services while simultaneously relaying the result back to the person who fell.

The core of behavior analysis is Taiwan GBA, developed by Academia Sinica in collaboration with Chung Yuan Christian University and National Tainan University. Its key concept is the Universal Action Space (UAS), inspired by DeepMind’s Kinetics-600 dataset—which collected videos of 600 types of human actions, each lasting about 5 to 10 seconds. The research team used these videos to establish an action feature space, enabling the model to learn correlations between different actions. In the future, new behaviors such as falling, waving, or attacking do not require building completely independent models; instead, they can be mapped into the existing action space. Liao Hung-yuan describes UAS as a "comprehensive sports stadium": different sports or behaviors can all be understood within this space.

However, human behavior cannot be judged from a single image; it requires understanding temporal changes. Two photos showing a person standing versus lying on the ground may appear as different states, but what truly matters is the process "from standing to falling." The research team utilizes Video Swin Transformer to analyze dynamic features in videos, identifying locations in the images related to actions—for instance, when playing tennis, the model focuses on the person’s hands, racket, and body movement areas—and converts these dynamic details into trainable behavior features, mapping them into the universal action space.

Sources :iThome

Recherche

UK Travel Booking Firm Holiday Extras Reports 32.5% Conversion Boost After Rebuilding App with OpenAI Codex

Prise rapide
2026-11-12 17:30 GMT+8

UK-based travel booking platform Holiday Extras stated in an OpenAI case study that it used OpenAI's coding tool, Codex, to rebuild its mobile application. According to the company, this process reduced development time from over six months to just two weeks.

Holiday Extras reported specific performance improvements: the main app's "open-to-book" conversion rate increased by 32.5%, and the app error rate dropped from 19% to 9%. Additionally, release frequency shifted from annual major updates to multiple weekly releases.

CEO Matthew Pack and VP of AI Innovation Liam Thompson noted that Codex enabled non-engineering staff to participate in software building, allowing customer insights to be directly translated into product features. For instance, after a customer left a one-star review due to a missing list view, the team quickly implemented a fix using Codex, leading the customer to change their rating to five stars.

It should be noted that all cited data are self-reported by Holiday Extras and have not been independently verified.

Sources :OpenAI

Recherche

Beijing Releases Three-Year Action Plan for Embodied AI Robots: Targets 50,000 PFlops Compute and 200,000 Units Annual Capacity

Sujet · 北京具身智能机器人方案Structurel
Vérifié 2026-11-12 18:44 GMT+8

On November 11, the Beijing Municipal Bureau of Economy and Information Technology issued the 'Action Plan for High-Quality Development of the Embodied Intelligent Robot Industry in Beijing (2026-2028)'. The plan positions the embodied AI robot industry as a key breakthrough for future industries and aims to drive high-quality development through six major projects.

Regarding infrastructure and technology, the plan proposes adding a dedicated compute cluster for embodied AI with a total scale exceeding 50,000 PFlops. It also sets a target to collect over 1 million hours of high-quality data from no fewer than 100 real-world scenarios. Support is provided for multiple technical routes, including Vision-Language-Action (VLA) models and World Models (WM), to enhance cognitive and execution capabilities.

In terms of industrial chain and manufacturing, the plan aims to form an annual complete machine production capacity of 200,000 units in Beijing by 2028. The Jing-Jin-Ji region's complete machine capacity is targeted to account for 25% of the national total, with a key component supporting rate exceeding 60%. Specific technical indicators for components like motor torque density and dexterous hand costs are also outlined.

For application promotion and safety governance, the plan intends to establish benchmark application scenarios in industrial, service, and special fields. It also prioritizes building a safety governance system covering the full lifecycle, including testing and certification centers and a 'one machine, one code' management system.

To ensure implementation, Beijing will formulate special support policies to increase funding for key technology breakthroughs and pilot construction, while providing support for urgently needed talents in areas such as household registration and housing.

Sources :jxj.beijing.gov.cn

Recherche

OpenAI Pledges $2 Million for AI Child Safety Research

Sujet · OpenAIPrise rapide
2026-11-12 08:00 GMT+8

OpenAI announced on November 12, 2026, that it has signed the Charter of Engagement for Safer AI for Children at the Paris Peace Forum and committed $2 million to support the scientific research network under the Coalition for Children’s Rights and Protection in the Age of AI.

This contribution supports shared measurement work: developing tools researchers can use to better understand and measure AI’s risks and opportunities for children, helping inform safer products and better policy. The charter sets out shared principles for safe, age-appropriate AI, alongside a research programme and safeguards for its independence. Research teams retain control over their conclusions, and findings will be published, including negative or inconclusive results.

This builds on OpenAI's call for global standards on youth AI safety and sustained collaboration among governments, researchers, civil society, and industry to translate shared evidence into practical safeguards.

Sources :OpenAI

Recherche

Google and Flight Attendants Union Reach Settlement to Scrub Spirit Airlines Data

Sujet · 精神航空数据脱敏协议Matériel
2026-11-12 18:15 GMT+8

According to WIRED, Google and the Association for Flight Attendants, the union representing crew members of the defunct Spirit Airlines, reached a groundbreaking settlement this week. The agreement addresses privacy concerns arising from Spirit Airlines' plan to sell its internal data assets to Google for AI model training during its liquidation process. Previously, Spirit sought to sell decades of data, including billions of emails and messages, for $10 million, but the union argued that these records contained significant personal employee information.

Under the revised terms, Spirit Airlines must hire a third-party data de-identification firm called Tonic AI to remove information potentially linked to individuals or consumers and ensure the anonymized data cannot be relinked to specific people. Google has committed to maintaining the data only in this newly de-identified form. The agreement explicitly excludes highly sensitive records, such as employee medical files, documents related to investigations and labor relations, and information about industry-wide safety management systems. A US aviation trade organization also intervened, warning that including such data could discourage employees from reporting safety issues through voluntary programs, which are critical to the industry's safety record.

The settlement includes an opt-out clause allowing Google to withdraw from the sale if the data-cleaning process reduces the asset value by more than 8 percent. Although viewed as a strong protection for employee privacy, the deal still requires judicial approval expected next week. Sara Nelson, president of the Association for Flight Attendants, stated that this outcome could serve as a template for other unions, which plan to demand similar data protection clauses in future labor agreements. The union is also working with state-level politicians to legislate worker data protections. This event highlights the growing tension between labor rights and the commercialization of private data amid the AI industry's surging demand for training materials.

Sources :WIRED

Recherche
Suivant →
Connexion / Inscription
LectureRechercheRadarCadre d'investissement

Monks also emphasized breadth as a hedge: this quarter it added growth sources not driven by AI, including banks benefiting from high US interest rates and light regulatory environments (Texas-based Cullen/Frost), energy companies profiting from tight supply (Cheniere Energy), and participation in the initial public offering (IPO) of Italian software integrator Bending Spoons. The reasons for selling WillScot and Stella-Jones were the lack of evidence of pricing power. This conscious effort to 'not become an AI infrastructure trust' is itself a form of managing single-theme risk.

In home care applications, the research team tested using Nanyang Technological University’s NTU RGB+D dataset, which contains 220 types of human actions. The team selected nine care-related actions for analysis. Liao Hung-yuan stated that in the dataset testing environment, the model’s recognition rate reaches approximately 80% to 90%. However, he also acknowledged that the biggest challenge in actual deployment is not laboratory data, but real-world data. Many crime datasets consist of simulated scenarios; for example, someone holding a knife might be acting, differing significantly from a real crime scene. For AI to be truly applied on-site, more real-world data accumulation is still required.

Regarding environmental understanding, Academia Sinica researcher Dr. Wang Chien-yao extended the YOLO v7 and YOLO v9 architectures to develop G-YOLO (Generalist YOLO). Traditional YOLO excels at real-time object detection, quickly answering "what things are in the picture," such as identifying people or cars in a street view. But for self-driving cars or robots, what is truly needed is understanding relationships between objects: if a group of people is crossing the road, AI must not only know "there are people" but also understand that this group is crossing the road, identify nearby buildings, and comprehend the relationships between people and surrounding objects. With G-YOLO added, the model can generate more complete scene descriptions, such as noting that approximately 15 people are crossing the road, with shops, buildings, and road facilities nearby.

Currently, Taiwan GBA can recognize 12 behaviors, including falling, fighting, and brandishing knives, covering application scenarios in home care, public safety, and campus security. G-YOLO continues to expand depth estimation and multi-view understanding capabilities, enabling AI to know not just "what was seen," but also "where objects are, how far away they are, and how they interact with each other."

Liao Hung-yuan believes that in the future, self-driving cars, robots, and smart care systems will require not only large language models but also a set of multimodal capabilities that can understand the real world. Large models allow AI to begin describing the world, but to truly enter the world, AI still needs to understand human actions, environmental changes, and the meaning behind events. This is the core direction of Academia Sinica’s investment in autonomous multimodal AI research and development.