AI BriefingMicrosoft Research news
The latest AI and developer news from Microsoft Research, with short summaries.
Visit website →MindTopo Reveals the Topological Spatial Reasoning Capabilities of VLMs
MindTopo measured the gap between VLMs' static topological recognition and interaction planning capabilities.
Introducing CARE-X: Implementing a Clinically Useful Radiology VLM via Auxiliary Supervision, Reward-Aligned Learning, and Tool-Augmented Measurement
CARE-X is presented as a chest X-ray VLM that combines generation and structured prediction.
Orchard: An Open-Source Framework for Scalable Agentic AI
Microsoft has released Orchard, an open-source framework for scaling agent training and evaluation.
Echoverse: A Deep, Evolving Environment for Computer-Use Agents
Microsoft has developed Echoverse, a highly precise training environment designed to maximize the performance of computer-use agents.
EvoLib: Turning Experience into Evolving Knowledge
EvoLib is a framework that lets AI agents teach themselves by converting past experiences into reusable skills and insights.
SymCrypt's Rust Cryptography Verification: From Standard to Code
Microsoft has unveiled a technique that leverages Rust, Aeneas, and Lean to verify the mathematical integrity of SymCrypt's cryptographic algorithms.
Aurora 1.5: Expanding the Open Foundation Model for Weather and Earth System Applications
Microsoft has open-sourced Aurora 1.5, an Earth system foundation model with expanded weather variables and ensemble forecasting capabilities.
Flint: A Visualization Language for the AI Era
Flint, an open-source visualization language that helps AI agents generate expressive charts, has been released.
SkillOpt: Agent Skills as Learnable Parameters
SkillOpt optimizes an agent's skills as learnable parameters without modifying model weights, increasing the reliability of its behavior.
Memora: A Balanced Memory Representation Combining Abstraction and Specificity
Microsoft has unveiled Memora, a memory framework that separates stored content from retrieval methods to enhance AI agents' ability to perform long-term tasks.
Understanding the Brain Through AI-Based Explanation and Experimentation
Microsoft Research unveiled the GCT framework, which uses LLMs to explain and verify how the brain processes language.
Talos: Scaling Rare Disease Diagnosis Through Automated Iterative Genomic Reanalysis
Developed in collaboration with Microsoft, Talos increases rare disease diagnosis rates and reduces the workload of medical staff through automated genomic reanalysis.
Ire Identifies Another LOTUSLITE Variant
Microsoft's autonomous malware classification agent, Project Ire, identified a LOTUSLITE variant with no existing indicators of compromise (IOC).
Data Formulator 0.7: AI-Powered Data Analysis for Enterprise Data
Data Formulator 0.7 is an open-source system that connects fragmented enterprise data and supports data analysis and visualization through AI agents.
Extending Human Intelligence Through AI
Modern AI does not replicate human intelligence, but operates by extending human cognitive and linguistic structures.
MagenticLite, MagenticBrain, Fara1.5: An Agentic Experience Optimized for Small Models
Microsoft has unveiled MagenticLite and Fara1.5, agentic systems designed for small models.
Vega: Zero-Knowledge Proofs for Digital Identity in the AI Era
Vega has implemented ZKPs that prove only facts from government-issued credentials without revealing the original documents.
Further Clarification on AI Delegation and Long-Horizon Task Reliability
Microsoft Research explained the reliability degradation of long-horizon AI delegation tasks.
mimalloc: A High-Performance Scalable Memory Allocator for Modern Times
mimalloc is an open-source allocator that serves as a drop-in replacement for malloc/free, optimized for large-scale multithreading.
GridSFM: A New Small Foundation Model for Power Grids
Microsoft has unveiled GridSFM, which predicts grid AC-OPF in milliseconds.