Apple reported a 38% year-over-year revenue increase in China, signaling a turnaround after a prolonged slump. This growth is attributed to strong iPhone sales and the popularity of services like Apple Music and iCloud. The company aims to capitalize on this momentum with targeted marketing strategies, particularly among younger consumers.
As Super Bowl LX approaches, the advertising landscape is heating up, with brands unveiling a wave of gameday ads. This year, expect a mix of celebrity endorsements and innovative campaigns aimed at capturing viewer attention during the Seattle Seahawks vs. New England Patriots matchup. Notably, brands are focusing on interactive elements to engage fans both on and off the field.
Reinforcement learning, particularly Proximal Policy Optimization (PPO), is crucial for fine-tuning Large Language Models (LLMs). However, PPO’s ratio clipping mechanism mismanages token updates, leading to inefficiencies. The proposed Divergence Proximal Policy Optimization (DPPO) replaces this with a principled divergence estimate, utilizing Binary and Top-K approximations to enhance stability and efficiency in training.
Researchers have developed FlatDINO, a variational autoencoder that compresses dense patch grids from models like DINOv2 into a one-dimensional sequence of 32 tokens, reducing dimensionality by 48x. On ImageNet 256x256, a DiT-XL using FlatDINO achieves a gFID of 1.80 while requiring 8x fewer FLOPs per forward pass and up to 4.5x less for training, indicating significant efficiency gains. Preliminary results suggest promising advancements in image generation models.
Researchers have developed the Continual Robotic Simulation Suite (CRoSS), a benchmark for continual reinforcement learning (CRL) using Gazebo-simulated robots. It features a two-wheeled robot and a seven-joint robotic arm, facilitating diverse tasks like line-following and goal-reaching. CRoSS offers kinematics-only variants for faster performance and includes a containerized setup for easy access and reproducibility, showcasing standard RL algorithms. This suite aims to enhance CRL research by providing a realistic and extensible testing environment.
Brian Hedden, PhD, has been named co-associate dean of the Social and Ethical Responsibilities of Computing (SERC) at MIT. This initiative, part of the MIT Schwarzman College of Computing, aims to integrate ethical considerations into technology development. Hedden's role will focus on fostering interdisciplinary dialogue and research on tech's societal implications.
Antonio Torralba from MIT has been selected for the 2025 cohort of the National Academy of Engineering. Renowned for his work in artificial intelligence and decision-making, Torralba’s research focuses on computer vision and machine learning. His appointment reflects a recognition of his contributions to the field, potentially enhancing MIT's influence in AI advancements.
Ubuntu 26.04 is set to launch for free in April, featuring enhanced security measures and performance improvements. Key updates include updated kernel support, refined user interfaces, and better integration with cloud services. This release aims to bolster user experience while addressing security vulnerabilities, making it a significant upgrade for both personal and enterprise users.
Mistral AI has unveiled a new suite of AI models designed to facilitate real-time multilingual conversation. This release aims to enhance communication across language barriers, potentially benefiting industries like customer service and global collaboration. The models are expected to improve accuracy and fluency in translations, marking a significant step in AI-driven language processing.
MIT senior Katie Spivakovsky has been awarded the 2026-27 Churchill Scholarship, allowing her to pursue an MPhil in biological sciences at the Wellcome Sanger Institute in the UK. This prestigious scholarship supports American students in advanced studies and research in the sciences, and Spivakovsky aims to enhance her research skills and contribute to the field.
At 3DEXPERIENCE World in Houston, NVIDIA's Jensen Huang and Dassault Systèmes' Pascal Daloz unveiled a new approach to industrial AI, emphasizing physics-based world models for systems design. This strategy aims to enhance simulation accuracy and efficiency in manufacturing, potentially transforming design processes across industries.
Intel's CEO announced at the Cisco AI Summit that the company will begin manufacturing a new type of chip, previously dominated by Nvidia. This move signals Intel's commitment to reclaiming its position in the competitive AI chip market, which is crucial as demand for AI technologies surges. The specifics of the chip design or production timeline were not disclosed, but this shift is seen as a strategic pivot for Intel to enhance its offerings in AI and data center solutions.