Category: World

Nokia Oyj has announced its first-quarter earnings, exceeding analysts’ expectations. The company’s focus has shifted toward artificial intelligence and cloud services, marking a significant change from its traditional hardware-centric model.

This transition has shown promising results, with adjusted profits reported at €0.12 per share. Analysts had anticipated earnings closer to €0.10. The company’s revenue grew, aided by increased demand for data center technologies and cloud solutions.

Nokia’s pivot has attracted attention and instilled confidence among investors. The integration of AI into its offerings has enhanced its competitive edge. As a result, the company is better positioned to capture market share in the evolving tech landscape.

The successful earnings report could signal a turnaround for Nokia after years of challenges. A strengthened focus on software and cloud services may lead to robust growth. This strategic shift could redefine the company’s future prospects and impact its standing in the industry.

Taiwan’s banking sector has relied heavily on global AI platforms for natural language processing. These systems, while powerful, often miss the nuances of local regulations and market practices. As financial technology evolves, this dependence has raised concerns among domestic banks.

In response, Taiwan’s financial authorities announced a plan to create a large language model specifically for local use. This initiative aims to empower local firms and enhance compliance with Taiwan’s regulatory landscape. The project represents a significant shift toward self-sufficiency in AI capabilities.

Tech firms, universities, and banks are collaborating on the development. By harnessing local expertise, they hope to create a more tailored AI solution that aligns with industry needs. The effort is supported by government funding and regulatory backing.

The anticipated AI model could reshape the competitive landscape for Taiwan’s banks. It aims to reduce reliance on foreign technology and enhance the effectiveness of local financial services. Ultimately, this initiative may foster innovation and bolster the region’s position in the global fintech arena.

Investors have grown accustomed to the significant premium attached to Taiwan Semiconductor Manufacturing Company’s (TSMC) American Depositary Receipts (ADR). This disparity typically reflects investor sentiment and market conditions, maintaining a palpable divide between TSMC’s Taiwanese shares and those traded in the US.

Recent developments, however, indicate a narrowing gap between TSMC’s Taiwanese stock and its US counterparts. A note from UBS Group AG highlights this shift, suggesting that the premium on TSMC’s ADR is diminishing, presenting new trading strategies for investors exposed to both markets.

The implications of this trend are substantial. As the gap closes, investors may find it easier to exploit price differences, facilitating trades that were previously less appealing. This could lead to increased trading volumes for TSMC’s ADR, with more market participants taking an interest in the stock.

The shift not only affects active traders but can also influence broader market perceptions of TSMC’s valuation. A more aligned pricing structure could enhance the appeal of TSMC among US investors, potentially increasing its market capitalization and influencing future investment strategies.

Large language models (LLMs) have become central to numerous applications, offering vast capabilities in natural language processing. However, the opacity surrounding their inference and training processes has often raised concerns among developers and researchers. Stakeholders have struggled to understand the true impacts of these models in real-world scenarios.

A recent paper has introduced a transparent screening framework aimed at addressing these challenges. This innovative approach enables users to estimate the inference and training impacts of LLMs, even with limited access to their inner workings. By transforming natural language application descriptions into bounded environmental estimates, the framework facilitates better evaluation and comparison of current market models.

The methodology avoids the pitfalls of relying on proprietary services by providing an auditable, source-linked approach. Its design is rooted in promoting comparability and reproducibility in the rapidly evolving landscape of large language models. Researchers now have a valuable tool that enhances understanding without compromising on transparency.

This development may shift how organizations evaluate AI technologies. As transparency becomes increasingly prioritized, businesses may feel more confident in adopting LLMs, ultimately driving innovation. The implications extend beyond individual models, potentially transforming industry standards for accountability in AI deployment.

Boston Consulting Group (BCG) has revealed that its artificial intelligence services generated 25% of its total revenue in 2025. This marks a significant change for the consulting firm, which traditionally relied heavily on its core management consulting services.

The firm has ramped up hiring efforts, bringing onboard more engineers and specialists to support a growing demand for AI expertise. These new team members are tasked with helping clients navigate the complexities of integrating AI technologies into their operations.

The results of this strategic pivot are clear. BCG’s clients are increasingly seeking innovative solutions that leverage AI for operational efficiency and competitive advantage. As such, the firm is positioned to continue expanding its influence in the market.

This shift has not only improved BCG’s financial outlook but also reflects broader trends in the business landscape. Companies across various sectors are recognizing the need for AI integration, driving demand for consulting services that can guide them through this transformation.

SK Hynix Inc. has announced a remarkable five-fold increase in its quarterly profit, signaling a robust demand for memory chips, particularly in the AI sector. This surge had positioned the company as a leader in the burgeoning market. Investors, however, are left questioning the sustainability of this growth.

The company’s impressive results stemmed from booming sales driven by the acceleration of AI technologies. Yet, market analysts note that this surge may not be indicative of a lasting trend. Many believe that the current demand may not sustain once the initial excitement around AI innovation fades.

Despite the positive financial report, investors responded with caution. SK Hynix’s stock did not rally as expected, reflecting concerns about potential volatility in the memory chip market. The debate surrounding whether a “supercycle” in demand is on the horizon or merely a temporary spike continues to cast a shadow over the company’s future prospects.

This uncertainty has led to a broader discussion within the tech industry regarding the longevity of AI-related investments. As companies navigate the implications of this shifting landscape, SK Hynix’s profit surge serves as a barometer for potential risks and rewards ahead. The skepticism surrounding future sales could redefine the company’s strategic approach in the coming quarters.

Tesla Inc. was poised to introduce its cutting-edge driver-assistance technology in China. This market has long been considered a key area for the company’s growth. Enthusiastic customers eagerly awaited new features touted to improve driving safety and convenience.

However, Tesla announced another delay in the rollout. Regulatory scrutiny in China has increased, with authorities expressing concerns over the safety and reliability of emerging automotive technologies. This caution reflects a broader trend among regulators to ensure that new innovations meet stringent safety standards.

In response, Tesla is reassessing its features and functionalities to comply with local regulations. The company has not set a new timeline for the release, leading to heightened speculation and disappointment among potential customers. Competitors are also watching closely, as they navigate the same regulatory landscape.

The postponement could impact Tesla’s sales figures in one of its largest markets. Analysts believe this delay may hinder the company’s competitive edge, especially as rivals launch their own advanced features. As the regulatory environment evolves, Tesla faces the challenge of balancing innovation with compliance.

PayPal’s Commerce Agent has relied on advanced fine-tuning to enhance its operational efficiency. Previously, the integration of the llama3.1-nemotron-nano-8B-v1 model led to notable performance improvements in transaction processing. However, the demand for even faster and more efficient solutions has increased.

The introduction of EAGLE3 marks a significant shift in optimization strategies. This new approach leverages speculative decoding to push the limits of throughput and latency without additional hardware costs. A recent empirical study compared EAGLE3’s performance on NVIDIA NIM using a variety of configurations.

Results demonstrated that using a gamma value of 3 delivered a 22-49% increase in throughput and reduced latency by 18-33%, with stable acceptance rates around 35.5%. Meanwhile, gamma=5 provided minimal benefits, indicating a saturation point in performance gains. LLM-as-Judge evaluations confirmed that the quality of outputs remained uncompromised during these enhancements.

This innovation enables significant cost reductions, allowing PayPal to match or even surpass previous benchmarks with only one H100 GPU—accomplishing up to a 50% reduction in GPU expenses. As a result, PayPal is well-positioned to offer quicker and more affordable services in an increasingly competitive landscape.

In traditional algorithm selection, engineers often relied on extensive domain knowledge to craft features that determine the best algorithm for a given problem. This reliance on hand-crafted instance features has constrained innovation and efficiency in diverse problem domains. However, a new approach known as ZeroFolio is challenging this norm.

ZeroFolio employs a feature-free methodology by leveraging pretrained text embeddings to eliminate the need for domain expertise. It transforms raw instance files into embeddings and applies a weighted k-nearest neighbors algorithm to select the optimal solution. This streamlined three-step process allows for broader application across various fields such as SAT and graph problems.

In testing, ZeroFolio achieved impressive results across 11 ASlib scenarios spanning seven different domains. It consistently outperformed conventional methods, including a random forest relying on hand-crafted features, in 10 of the scenarios and in all 11 when using a two-seed voting technique. The findings underscore the potential of embedding models to create competitive advantages without prior training.

The implications of this method are significant. By reducing the dependence on domain knowledge, ZeroFolio makes advanced algorithms more accessible to practitioners unfamiliar with specific areas. This democratization of algorithm selection could accelerate advancements in AI and problem-solving across various domains, fostering innovation and collaboration.

Geometric tempering has traditionally been a tool for improving the sampling of probability distributions. Researchers have relied on it to minimize the Kullback–Leibler divergence from target distributions. The standard approach is well-understood; however, recent explorations have revealed deeper complexities.

Recent findings introduce a sequence of moving targets defined by geometric tempering. This new framework alters the dynamics of sampling through Wasserstein and Fisher–Rao gradient flows. Notably, researchers observed that convergence can occur exponentially in continuous time, offering fresh insights into optimization processes.

Extensive analysis demonstrated that while popular time discretizations exist, their convergence properties can vary significantly. In particular, findings indicated that using a geometric mixture of distributions does not accelerate convergence speeds. This stands in contrast to conventional expectations in both continuous and discrete settings.

The implications of this study extend to the development of adaptive tempering schedules, which can enhance gradient flow structures. These new strategies may redefine how researchers approach probability distributions in machine learning. The landscape of gradient flow dynamics is shifting, presenting both challenges and opportunities for future work.