Alibaba's Qwen3-TTS repository on GitHub has accumulated over 14,000 stars and was updated today. It is an open text-to-speech model release in the Qwen family.
MIT Technology Review reports that Texas, recently the epicenter of AI data center construction, may see that boom cool. The piece ties the shift to broader infrastructure economics.
Protesters broke into a Microsoft data center site in Amsterdam, per The Verge. The incident adds to a wave of local opposition to AI infrastructure.
OpenAI is reportedly retreating from NDAs attached to its data center projects. The move follows scrutiny over infrastructure transparency.
A Hugging Face paper applies sparse autoencoders to Mixture-of-Experts models, finding experts show fine-grained human-interpretable specialization. This bears on MoE interpretability and routing.
A new arXiv paper, Scientist World, proposes growing a simulated research environment for AI-scientist agents. It appeared on both arXiv and Hugging Face papers.
Physicl raised $10M to build digital environments realistic enough for robot training. The funding targets simulation fidelity for embodied AI.
Underdog demonstrated a 32MB speech model, emphasizing extreme size reduction. The model is positioned for edge deployment.
Amazon now lets users text Alexa to build shopping lists and place orders, described as a new shopping agent capability. This follows Amazon blocking a service called Muse.
AWS documented a serverless MCP server pattern for AI agent access to data lakes on ECS Fargate. The reference architecture addresses security and scaling for agent-data integrations.
AWS published documentation on governing AI agent access using Okta Cross App Access with Bedrock AgentCore. The post covers identity and access patterns for production agents.
Palo Alto Networks is reportedly negotiating to buy AI agent security startup Clutch for roughly $500M. The deal would fold agent-security tooling into a major network security vendor.
TechCrunch reports Anthropic is approaching an IPO while Ramp payment data indicates enterprise AI spending is flattening. The juxtaposition raises questions about growth expectations at the offering.
NVIDIA published Cosmos 3 open models aimed at interactive world simulation, available on Hugging Face. The release targets embodied AI and simulation-driven training.
OpenAI announced GPT-6 Cyber, a frontier model specialized for cybersecurity agent workflows. The post frames it as advancing aligned cybersecurity agents, implying safety-trained offensive/defensive capabilities.