News Hub

Moving AI inference off robots onto external systems improves task success and efficiency for physical AI workloads.

GigaPath-Flash and GigaTIME-Flash reduce computational demands for pathology foundation models.

MindTopo is a benchmark for testing vision language models' understanding of topological relationships and spatial reasoning.

CARE-X combines auxiliary supervision, reward-aligned learning, and tool-augmented measurement for chest X-ray interpretation.

Orchard is an open-source framework for training and evaluating AI agents across task types with reusable infrastructure.

Microsoft introduces Echoverse, evolving environments for training computer-use agents on multi-step workflows.

Microsoft presents EvoLib to help LLMs develop reusable skills and adapt across tasks post-deployment.

New method verifies Rust cryptography in SymCrypt while preserving speed and adaptability during implementation and evolution.

Aurora 1.5 adds 22 variables, hourly resolution, and probabilistic forecasting to the Aurora weather foundation model.

Flint is an open-source visualization language for AI agents to create expressive charts.

SkillOpt converts manual agent skill editing into a trainable parameter optimization process without changing model weights.

Memora memory system separates storage from retrieval to scale AI agent context beyond reload and retrieval inefficiencies.