OpenAI, Hugging Face, Anthropic, Nvidia: аудит скорости

<!doctype html> OpenAI, Hugging Face, Anthropic, Nvidia: аудит скорости OpenAI, Hugging Face, Anthropic, Nvidia: аудит скорости

В этом выпуске: темп развития AI, агентные инциденты, криптоанализ, вычислительный капитал, открытые веса, маршрутизация моделей, управляемые агенты, научные вычисления, формальная верификация, обучение, геоинфраструктура и видеомиры.

Оригинальные материалы

• Latent Space: frontier labs, Pace AI development and Hugging Face cyber incident
• Simon Willison: Anatomy of a Frontier Lab Agent Intrusion
• The Decoder: Anthropic Claude Mythos and cryptographic vulnerabilities
• The Decoder: Amazon scales back Nova and bets on Frontier Model Research
• The Decoder: Nvidia invests in SSI
• The Decoder: Dario Amodei on open-weight risks
• BBC: chip stocks slide as AI jitters rattle investors
• LearnVector: Andrew Ng’s one-to-one learning experiences
• LeanScreen: Lean Verification
• OpenAI: Scientific computing in the age of agentic AI
• Hugging Face / AllenAI: The OlmoEarth Platform
• Google: Gemini API Managed Agents, 3.6 Flash, hooks and triggers
• MarkTechPost: Fireworks Nexus routing and cost-control layer
• WorkOS: MCP Server Empowers Agents
• Pragmatic Engineer: How building software is changing at Anthropic
• Hugging Face Papers: Wonder, Video World Model Done Better