NVIDIA’s Nemotron 3 family introduces a hybrid Mamba‑Transformer + MoE backbone built for efficient, accurate agentic reasoning at million‑token scale—openly licensed and deployable across common platforms.

Read more

Microsoft Research proposes PlugMem, a memory layer that converts raw agent interactions into compact facts and skills organized in a retrievable graph—cutting redundancy, reducing tokens, and improving decision relevance.

Read more

OpenAI long advertised strict limits on military and surveillance uses. Yet Microsoft, its chief partner, integrated OpenAI models into Azure Government for defense and intelligence customers under U.S. compliance regimes—enabling testing and, later, top-secret deployments.

Read more

Grammarly’s rebrand to Superhuman signals a shift from grammar checks to cross-app AI agents. Here’s what that means for author-style feedback, the ethics debate, and how businesses should evaluate the risks and opportunities.

Read more

MIT researchers embedded a tabular foundation model into a Bayesian optimization loop, reporting 10–100× faster convergence on engineering-style benchmarks. The approach could make high-dimensional design tasks more accessible through spreadsheet-native interfaces.

Read more

NVIDIA’s GTC returns to San Jose on March 16–19, 2026, with Jensen Huang’s keynote expected to spotlight the emerging age of AI, from physical AI to agentic systems. Here’s what leaders, builders, and investors should know—plus how to attend.

Read more