A mixture of experts, or MoE, is a model architecture that contains multiple specialized subnetworks and a routing system that selects which experts handle each input. Only a subset may be activated for a given token or example, allowing the model to contain many parameters without using all of them for every calculation. This can expand capacity while controlling inference cost. Training remains complex because routing must distribute work effectively, avoid overloading a few experts, and maintain communication across hardware. An MoE model is not automatically interpretable simply because its components are called experts; their learned specializations may be unstable or difficult to describe and must be evaluated empirically.
SoftBank has secured a $40 billion bridge loan to deepen its investment in OpenAI and accelerate its broader AI strategy.
AI-driven bots now generate more internet traffic than humans, according to a new report, highlighting the growing impact of automated systems online.
New research shows that telling AI models they are “experts” can reduce accuracy in tasks like coding and math, while improving safety and alignment outcomes.
Meta introduced four new in-house MTIA chips designed for AI training and inference as the company accelerates data center expansion. The chips aim to improve performance and reduce reliance on external hardware suppliers.
Google has rolled out Canvas in AI Mode to all U.S. users in English, enabling people to create documents, dashboards, and interactive tools directly within Search. The feature provides a dynamic workspace for planning projects and building simple applications.
Andrej Karpathy argues AI coding agents now fundamentally change programming workflows, enabling long-running autonomous tasks and redefining software engineering.
MSCI introduces AI connectors enabling access to its data via MSCI ONE, ChatGPT, and Claude, debuting IndexAI Insights to deliver conversational index analytics powered by large language models.
Anthropic unveils a revised Responsible Scaling Policy with a Frontier Safety Roadmap, regular Risk Reports, and clearer separation between company commitments and industry recommendations.
ŌURA unveils a proprietary large language model for women’s health, combining clinical research and biometric data to deliver personalized, privacy-first AI guidance through Oura Advisor.
Anthropic reveals industrial-scale campaigns by DeepSeek, Moonshot, and MiniMax to extract Claude’s capabilities via fraudulent accounts, highlighting national security and AI safety risks.