NVIDIA’s SIGGRAPH update shifts physical AI from cloud demos toward edge deployment. The package includes the 4B Cosmos 3 Edge world model, a Synthetic Video Detector NIM microservice, and a DGX Station agent stack built around Nemotron 3 Ultra.
The policy fight is no longer just about model benchmarks. Axios reports that U.S. officials have revisited tools such as Entity List threats, security advisories, procurement pressure, and hosting liability rules as cheaper Chinese open-weight models gain enterprise traction.
Databricks’ Summit recap compresses a broad enterprise AI roadmap into five minutes. The product list includes Genie One, Ontology, App Builder, ZeroOps, LTAP, Unity AI Gateway, Omnigent and CustomerLake.
Long-document OCR is bottlenecked by page chunking and growing KV cache. A widely shared post says Baidu’s Unlimited-OCR uses 3B total parameters, 500M active parameters, and a 32K context window to read 40-page documents in one pass.
GitHub’s security cleanup shows that the hard part is not finding secrets, but routing real risk to the right owners. The company says it cleared 20,000-plus alerts across 15,000 repositories in nine months.
OpenAI says nearly 9 in 10 teens on ChatGPT use it weekly for learning, information, skill-building, or productivity. The new safety push lets parents enable Study Mode by default for linked teen accounts and expands notifications for serious policy violations.
Video analytics development is moving from hand-built pipeline wiring toward natural-language instructions plus coding agents. DeepStream 9.1 adds 13 agentic skills and JetPack 7.2 support.
Google Research links diffusion-model novelty to score smoothing, not a mysterious creative spark. The ICLR 2026 paper and released code give the memorization debate a sharper mechanism: regularized networks interpolate between training samples instead of collapsing onto them.
Dependabot will now wait until a new package release has been on its registry for at least three days before opening a version-update pull request. Security updates still open immediately, and teams can tune or opt out through cooldown configuration.
GitHub code scanning can now show AI-powered security detections directly on pull requests. The preview extends coverage beyond languages and frameworks supported by CodeQL, while requiring GitHub Code Security, a Copilot license, and AI credits.
Verified U.S. K-12 teachers can get a full year of Claude access at no cost. The important shift is that Anthropic is packaging premium Claude, teaching skills, standards-aligned curriculum context, Claude Code, and Cowork into one education workflow.
NVIDIA showed Cosmos 3 Nano rising from 54.41% zero-shot accuracy to 93.35% after LoRA and TAO AutoML on a traffic safety video QA task. The result frames agent-run post-training as a practical physical AI workflow.