Nvidia Posts First $200 Billion Year on Back of AI Investment Boom In a landmark achievement, Nvidia has reported its first-ever $200 billion year, driven by the surging demand for its AI-powered technologies across various industries. This milestone underscores the company's pivotal role in fueling the artificial intelligence (AI) revolution that is transforming the global economy. According to the company's latest financial results, Nvidia's revenue surged 73% year-over-year to a record $68.1 billion in the fourth quarter of 2025. Its profit nearly doubled to $43 billion, or $1.76 per share, far exceeding analyst projections. "AI is just going to be everywhere. So we have plenty of runway, lots and lots of growth ahead of us," Nvidia CEO Jensen Huang said in an interview. The AI boom has become a significant driver of the U.S. economy, accounting for nearly a fifth of the 4.2% year-over-year increase in GDP in the fourth quarter of 2025, according to an analysis by Pantheon Macroeconomics. Additionally, the "wealth effect" from the surge in tech stock prices, including Nvidia's, contributed an estimated 0.3 percentage points to GDP growth during the same period. Nvidia's success is closely tied to the Indian government's ambitious IndiaAI Mission, which is investing over $1 billion to bolster the nation's AI ecosystem. The company is collaborating with leading cloud providers like Yotta, L&T, and E2E Networks to build out India's AI infrastructure, including the deployment of tens of thousands of Nvidia GPUs. Beyond India, Nvidia is also playing a crucial role in securing critical infrastructure worldwide. The company is working with cybersecurity providers like Akamai, Forescout, Palo Alto Networks, and Xage Security to bring its accelerated computing and AI capabilities to operational technology (OT) and industrial control systems (ICS) environments, enhancing real-time threat detection and response. However, the AI boom has also raised concerns about its potential impact on the job market. Nvidia's CEO acknowledged that the "concerns about China relying on American technology to advance their AI industry are just poorly placed," as China has its own technological capabilities in this domain. As the AI revolution continues to unfold, Nvidia's record-breaking performance underscores its pivotal position at the forefront of this transformative technology. The company's ability to capitalize on the growing demand for AI-powered solutions across diverse industries has solidified its status as a key driver of the global economy's digital transformation.
Google Unveils AI-Driven Grid Optimization Technology In a significant development, tech giant Google has partnered with grid technology company CTC Global to launch a new AI-driven product aimed at improving the efficiency and capacity of electricity grids. The technology, which is currently in the pilot stage, utilizes fiber-optic sensors to gather real-time data on power line activity and strain. This data is then processed using Google's software, including Google Earth Engine and BigQuery, and fed into Tapestry, a grid modeling project by Google's parent company, Alphabet's moonshot lab X. According to CTC CEO J.D. Sitton, this system allows utilities to push up to 120% more power through existing lines, at a lower cost than building new power generation stations. "Efforts like this are becoming more common as a growing number of tech and utility companies realize they won't be able to build new infrastructure fast enough to keep up with rapidly increasing demand," the Semafor report states. Alongside its grid optimization efforts, Google has also made significant strides in reducing the environmental impact of its AI operations. According to a report by TechReport, the company has achieved a 33-fold decrease in energy consumption per AI query and a 44-fold reduction in emissions over the past year. Google attributes these improvements to advanced model architectures, custom-built hardware, optimized inference algorithms, and highly efficient data centers. However, the report also highlights the potential risks posed by the increasing demand for AI, which could strain global energy and water resources. Google emphasizes the importance of transparent, comprehensive frameworks for measuring AI efficiency to ensure sustainability, advocating for industry-wide adoption of such standards. In a separate development, Google has also expanded its AI creative studio, Flow, with new features and integrations. The platform now includes Google's image generation experiments, Whisk and ImageFX, and allows users to generate images and use them as the basis for videos. Additionally, Flow has introduced a lasso tool for targeted image editing, flexible media management, and tools for extending clips and controlling camera movements. Furthermore, Google is preparing a major overhaul of the Build feature in its AI Studio, transforming it from a simple app prototyping tool into a more capable platform for building full-stack, multiplayer applications. The update is expected to include a new authentication layer that will integrate with Firebase, as well as expanded framework support beyond the current JavaScript and React options. Overall, Google's latest initiatives in grid optimization, AI efficiency, and creative studio enhancements demonstrate the company's commitment to leveraging its technological expertise to address pressing energy and environmental challenges, while also expanding the capabilities of its AI-powered tools and platforms.
Distilling the tool-using capabilities of large language models (LLMs) into smaller, more efficient small language models (SLMs) is a key challenge for their practical application. The predominant approach, supervised fine-tuning (SFT), suffers from poor generalization as it trains models to imitate a static set of teacher trajectories rather than learn a robust methodology. While reinforcement learning (RL) offers an alternative, the standard RL using sparse rewards fails to effectively guide SLMs, causing them to struggle with inefficient exploration and adopt suboptimal strategies. To address these distinct challenges, we propose MENTOR, a framework that synergistically combines RL with teacher-guided distillation. Instead of simple imitation, MENTOR employs an RL-based process to learn a more generalizable policy through exploration. In addition, to solve the problem of reward sparsity, it uses a teacher's reference trajectory to construct a dense, composite teacher-guided reward that provides fine-grained guidance. Extensive experiments demonstrate that MENTOR significantly improves the cross-domain generalization and strategic competence of SLMs compared to both SFT and standard sparse-reward RL baselines.
Gradient-based data attribution methods, such as influence functions, are critical for understanding the impact of individual training samples without requiring repeated model retraining. However, their scalability is often limited by the high computational and memory costs associated with per-sample gradient computation. In this work, we propose GraSS, a novel gradient compression algorithm and its variants FactGraSS for linear layers specifically, that explicitly leverage the inherent sparsity of per-sample gradients to achieve sub-linear space and time complexity. Extensive experiments demonstrate the effectiveness of our approach, achieving substantial speedups while preserving data influence fidelity. In particular, FactGraSS achieves up to 165% faster throughput on billion-scale models compared to the previous state-of-the-art baselines. Our code is publicly available at https://github.com/TRAIS-Lab/GraSS.
Large language models (LLMs) have demonstrated promising performance in generating diagnostic conclusions from imaging findings, thereby supporting radiology reporting, trainee education, and quality control. However, systematic guidance on how to optimize prompt design across different clinical contexts remains underexplored. Moreover, a comprehensive and standardized framework for assessing the trustworthiness of LLM-generated radiology reports is yet to be established. This study aims to enhance the trustworthiness of LLM-generated liver MRI reports by introducing a Multi-Dimensional Credibility Assessment (MDCA) framework and providing guidance on institution-specific prompt optimization. The proposed framework is applied to evaluate and compare the performance of several advanced LLMs, including Kimi-K2-Instruct-0905, Qwen3-235B-A22B-Instruct-2507, DeepSeek-V3, and ByteDance-Seed-OSS-36B-Instruct, using the SiliconFlow platform.
