Latest AI News

DAWN: Dependency-Aware Fast Inference for Diffusion LLMs

The article introduces DAWN, a new method for improving inference speed in diffusion large language models (dLLMs) without sacrificing output quality. DAWN addresses the inefficiencies of traditional parallel decoding by modeling inter-token dependencies, allowing for more reliable token unmasking. Experimental results show DAWN enhances inference speed by 1.80-8.06x compared to existing methods while maintaining generation quality. The code is available at GitHub for implementation.

arXiv

21 days ago

Spotify’s About the Song offers context and trivia about your favorite tracks

A new app feature allows users to explore the stories behind their favorite songs directly within the platform. This enhancement aims to enrich the listening experience by providing background on artists, songwriting processes, and thematic elements without requiring users to navigate away. The feature's launch is expected to deepen engagement and attract more users seeking a more immersive musical experience.

The Verge

21 days ago

3 Ways NVFP4 Accelerates AI Training and Inference

NVIDIA has unveiled its latest AI model architecture, designed to address the escalating computational demands for training and inference that exceed Moore's Law. The new system, featuring advanced GPUs and optimized software, aims to enhance efficiency and performance, catering to the growing needs of AI applications across industries. This innovation is critical for developers seeking to leverage AI capabilities without prohibitive costs or resource constraints.

Nvidia.com

21 days ago

Can China’s No. 2 automaker make it in America?

Geely, China's second-largest automaker, is exploring the possibility of manufacturing vehicles in the U.S. However, they must navigate strict cybersecurity regulations that govern automotive software. Compliance with these requirements will be crucial for their success in the competitive American market.

The Verge

22 days ago

OpenAI launches new agentic coding model only minutes after Anthropic drops its own | TechCrunch

On Monday, OpenAI unveiled Codex, a new coding tool aimed at software developers, alongside the launch of GPT-5.3 Codex, which enhances Codex’s capabilities. This updated model promises improved code generation and debugging features, allowing developers to streamline their workflows. OpenAI's advancements could significantly impact software development efficiency, though specifics on performance metrics are pending.

TechCrunch

22 days ago

Pseudo-Invertible Neural Networks

Researchers have introduced Surjective Pseudo-invertible Neural Networks (SPNN), a novel architecture that extends the Moore-Penrose Pseudo-inverse to nonlinear systems, particularly in neural networks. This includes a method called Non-Linear Back-Projection (NLBP), which ensures consistency in solving nonlinear inverse problems. By applying diffusion-based techniques, SPNNs can tackle various non-linear degradations, from optical distortions to semantic classification, enabling zero-shot inversion and precise control over generative outputs without requiring retraining.

arXiv

22 days ago

DFlash: Block Diffusion for Flash Speculative Decoding

DFlash introduces a novel speculative decoding framework utilizing a lightweight block diffusion model for parallel token generation, enhancing efficiency in large language models. By producing draft tokens in one forward pass and conditioning on context from the target model, DFlash achieves over 6x acceleration and up to 2.5x faster speeds than EAGLE-3, improving inference quality and GPU utilization.

arXiv

22 days ago

Can vision language models learn intuitive physics from interaction?

Recent research indicates that pre-trained vision-language models struggle with physical world intuitions. Although supervised fine-tuning enhances performance on simple tasks, it doesn't yield robust, generalizable physical rules. Experiments using reinforcement learning for interaction-based training improved task-specific performance but failed to ensure generalization across related tasks, even when visual and physical similarities existed.

arXiv

22 days ago

GPT-5.3-Codex

OpenAI has unveiled GPT-5.3-Codex, an advanced coding model designed to enhance coding capabilities significantly. This model reportedly improves performance in generating code and understanding complex programming tasks. It offers new features for developers, potentially streamlining workflows and increasing productivity in software development.

Openai.com

22 days ago

Your ExpressVPN is getting three huge upgrades - including a private AI assistant

ExpressVPN has introduced two new features: ExpressKeys, a password manager, and ExpressMailGuard, an email security tool. ExpressKeys securely generates and stores passwords, while ExpressMailGuard protects users from phishing attacks. These additions aim to enhance user security and privacy in an increasingly digital landscape.

ZDNet

23 days ago

ExpressVPN’s New Privacy-Focused AI and Email Protection Features Could Be Game-Changers

ExpressVPN, CNET's top-rated VPN, has rolled out significant updates aimed at enhancing user privacy. Key features include a new Lightway protocol for faster speeds, an expanded server network, and improved support for obfuscation techniques to bypass censorship. These updates are designed to bolster security and enhance user experience, making ExpressVPN a more versatile tool for privacy-conscious consumers.

CNET

23 days ago

Amazon's Alexa Plus Confounds Chris Hemsworth in Super Bowl Ad as the AI Tool Launches

Amazon is set to unveil the Alexa Plus AI to the public, following a year of early access. The official preview will occur during a Super Bowl ad featuring Chris Hemsworth. This marks a significant step for Amazon, as it aims to enhance user interaction with Alexa through advanced AI capabilities. Expect more details on features and user experience soon.

CNET

23 days ago