# Microsoft Research

Explore research at Microsoft, a site featuring the impact of research along with publications, products, downloads, and research careers.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## Can AI help build a healthier future?

DevFeed: [Can AI help build a healthier future?](<https://devfeed.tech/articles/can-ai-help-build-a-healthier-future-139417.md>)

Original publisher: [Read original article](<https://www.microsoft.com/en-us/research/podcast/can-ai-help-build-a-healthier-future/>)

Author: Doug Burger, Ava Amini, Eric Horvitz, Dr. Eric Topol

Published: 2026-11-17T16:52:17Z

Content type: article

Language: en

Sources: [Microsoft Research](<https://devfeed.tech/sources/microsoft-research.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Predictive AI](<https://devfeed.tech/topics/predictive-ai.md>), [AI Chat](<https://devfeed.tech/topics/ai-chat.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [2016](<https://devfeed.tech/tags/2016.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-help](<https://devfeed.tech/tags/ai-help.md>), [ava](<https://devfeed.tech/tags/ava.md>), [build](<https://devfeed.tech/tags/build.md>), [can](<https://devfeed.tech/tags/can.md>), [explore](<https://devfeed.tech/tags/explore.md>), [first](<https://devfeed.tech/tags/first.md>), [future](<https://devfeed.tech/tags/future.md>), [help](<https://devfeed.tech/tags/help.md>), [lead](<https://devfeed.tech/tags/lead.md>), [manager](<https://devfeed.tech/tags/manager.md>), [medicine](<https://devfeed.tech/tags/medicine.md>), [microsoft](<https://devfeed.tech/tags/microsoft.md>), [microsoft-research](<https://devfeed.tech/tags/microsoft-research.md>), [microsoft-research-podcast](<https://devfeed.tech/tags/microsoft-research-podcast.md>), [more](<https://devfeed.tech/tags/more.md>), [post](<https://devfeed.tech/tags/post.md>), [questions](<https://devfeed.tech/tags/questions.md>), [research](<https://devfeed.tech/tags/research.md>)

### AI overview

AI could improve diagnosis, guide biological research and support prevention, but the conversation stresses that research must be implemented in care and validated with better data. Eric Horvitz describes testing language models through sequential diagnosis: asking questions, choosing tests and deciding when to reach a diagnosis under uncertainty. Eric Topol points to AI's established potential in medical imaging and says adoption of findings from studies has lagged, citing cost pressures and slow health-system change. Ava Amini sees models helping prioritize experiments and reduce lab work, while emphasizing that real-world experiments remain necessary. The guests identify richer, long-term patient data and functional biological measurements as key gaps; they also caution that exact, complete models of biological systems may be unattainable or unnecessary for useful discoveries. They discuss possible gains in clinical workflows, drug discovery and prevention, while framing future breakthroughs as expectations rather than established outcomes.

### Source excerpt

What can AI enable in medicine, and how do we get there? Microsoft Research lead Doug Burger, Microsoft Principal Research Manager Ava Amini, Microsoft Chief Scientific Officer Eric Horvitz, and physician-scientist Eric Topol of Scripps Research explore these questions and more. The post Can AI help build a healthier future? appeared first on Microsoft Research.

## TokenEconomizer: Learning from Experience to Make AI Coding Agents More Efficient

DevFeed: [TokenEconomizer: Learning from Experience to Make AI Coding Agents More Efficient](<https://devfeed.tech/articles/tokeneconomizer-learning-from-experience-to-make-ai-coding-agents-more-efficient-136424.md>)

Original publisher: [Read original article](<https://www.microsoft.com/en-us/research/blog/tokeneconomizer-learning-from-experience-to-make-ai-coding-agents-more-efficient/>)

Author: Aaron J. Li, Weijia Xu, Chandan Singh, Jeevana Priya Inala, Bin Yu, Ion Stoica, Michel Galley, Jianfeng Gao

Published: 2026-11-16T17:01:35Z

Content type: article

Language: en

Sources: [Microsoft Research](<https://devfeed.tech/sources/microsoft-research.md>)

Topics: [agentic-coding](<https://devfeed.tech/topics/agentic-coding.md>), [AI-assisted coding](<https://devfeed.tech/topics/ai-assisted-coding.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>)

Tags: [adapting](<https://devfeed.tech/tags/adapting.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-coding](<https://devfeed.tech/tags/ai-coding.md>), [ai-coding-agents](<https://devfeed.tech/tags/ai-coding-agents.md>), [coding](<https://devfeed.tech/tags/coding.md>), [coding-agents](<https://devfeed.tech/tags/coding-agents.md>), [costs](<https://devfeed.tech/tags/costs.md>), [experience](<https://devfeed.tech/tags/experience.md>), [learn](<https://devfeed.tech/tags/learn.md>), [learning](<https://devfeed.tech/tags/learning.md>), [microsoft](<https://devfeed.tech/tags/microsoft.md>), [microsoft-research](<https://devfeed.tech/tags/microsoft-research.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [plugin](<https://devfeed.tech/tags/plugin.md>), [research](<https://devfeed.tech/tags/research.md>), [research-blog](<https://devfeed.tech/tags/research-blog.md>), [sessions](<https://devfeed.tech/tags/sessions.md>), [tasks](<https://devfeed.tech/tags/tasks.md>), [tokens](<https://devfeed.tech/tags/tokens.md>), [waste](<https://devfeed.tech/tags/waste.md>)

### AI overview

TokenEconomizer uses repository session history to reduce coding-agent costs without changing the agent or its model. It adds confirmed recurring patterns as short startup notes, prunes tools unlikely to be needed while retaining command and file-editing tools, and uses a judge model to detect repeated failures and suggest another approach. Across 277 SWE-Bench tasks in six repositories and five agent/model combinations, costs fell 18.4-30.9% with task success broadly stable; savings included the plugin's own model calls. Early tasks could cost more while it learns, but cumulative costs were below the default by task six in every repository. The excerpt also reports savings on sequential web development and security auditing, where each task used a new repository. The plugin keeps learned data on the developer's machine and requires no API key of its own.

### Source excerpt

AI coding agents can waste time and tokens relearning the same lessons. TokenEconomizer is an open-source plugin that builds on past sessions to cut costs while maintaining task success. It's a path toward coding agents that learn not only how to solve tasks, but also how to spend computation more effectively. The post TokenEconomizer: Learning from Experience to Make AI Coding Agents More Efficient appeared first on Microsoft Research.

## Microsoft Research Focus - Edition #98

DevFeed: [Microsoft Research Focus - Edition #98](<https://devfeed.tech/articles/microsoft-research-focus-edition-98-131231.md>)

Original publisher: [Read original article](<https://www.microsoft.com/en-us/research/podcast/microsoft-research-focus-edition-98/>)

Author: Anson Ho

Published: 2026-11-12T21:31:57Z

Content type: news

Language: en

Sources: [Microsoft Research](<https://devfeed.tech/sources/microsoft-research.md>)

Topics: [AI Research](<https://devfeed.tech/topics/ai-research.md>), [AI Context Management](<https://devfeed.tech/topics/ai-context-management.md>), [Microsoft](<https://devfeed.tech/topics/microsoft.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-coding-agents](<https://devfeed.tech/tags/ai-coding-agents.md>), [algorithms](<https://devfeed.tech/tags/algorithms.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [bottlenecks](<https://devfeed.tech/tags/bottlenecks.md>), [breakthrough](<https://devfeed.tech/tags/breakthrough.md>), [compaction](<https://devfeed.tech/tags/compaction.md>), [compression](<https://devfeed.tech/tags/compression.md>), [computer-vision](<https://devfeed.tech/tags/computer-vision.md>), [computing](<https://devfeed.tech/tags/computing.md>), [context](<https://devfeed.tech/tags/context.md>), [context-compaction](<https://devfeed.tech/tags/context-compaction.md>), [costs](<https://devfeed.tech/tags/costs.md>), [efficiency](<https://devfeed.tech/tags/efficiency.md>), [efficiently](<https://devfeed.tech/tags/efficiently.md>), [mathematics](<https://devfeed.tech/tags/mathematics.md>), [microsoft-research](<https://devfeed.tech/tags/microsoft-research.md>), [microsoft-research-podcast](<https://devfeed.tech/tags/microsoft-research-podcast.md>), [research](<https://devfeed.tech/tags/research.md>)

### AI overview

The edition highlights several Microsoft research developments. Researchers resolved the 31-year-old KLS conjecture, establishing that simple flat cuts locate bottlenecks in log-concave distributions within a fixed factor regardless of dimension; the results were independently validated and the proof formally verified with computer assistance. For long coding-agent sessions, a study of 13.5 million sessions found that fewer than 8% needed context compaction but they accounted for 44% of AI processing costs. The researchers introduced the COMPACTIONBENCH benchmark and APEC, which prepares summaries in the background; evaluations reported coding performance gains, cost reductions of up to 25.4%, and execution time reductions of 9.6%. A Latent Core Tokenizer (LCT) identifies reusable word components before building a multilingual vocabulary. Across 104 languages it yielded more linguistically meaningful tokens at comparable efficiency and improved results on four multilingual benchmarks. The excerpt also previews AI-assisted universal translation while noting challenges remain.

### Source excerpt

What's new in research at Microsoft? Host Anson Ho explores the latest, including AI-powered universal translation, a breakthrough in a 31-year-old mathematics challenge, and advances making AI coding agents faster and more efficient. The post Microsoft Research Focus - Edition #98 appeared first on Microsoft Research.

## Lazare, veni foras: Reprogramming Dormant Neurons to Backdoor AI Agents

DevFeed: [Lazare, veni foras: Reprogramming Dormant Neurons to Backdoor AI Agents](<https://devfeed.tech/articles/lazare-veni-foras-reprogramming-dormant-neurons-to-backdoor-ai-agents-127781.md>)

Original publisher: [Read original article](<https://www.microsoft.com/en-us/research/blog/lazare-veni-foras-reprogramming-dormant-neurons-to-backdoor-ai-agents/>)

Author: Dimitris Papailiopoulos

Published: 2026-11-10T20:52:56Z

Content type: news

Language: en

Sources: [Microsoft Research](<https://devfeed.tech/sources/microsoft-research.md>)

Topics: [Neural Network](<https://devfeed.tech/topics/neural-network.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-security](<https://devfeed.tech/tags/ai-security.md>), [backdoor](<https://devfeed.tech/tags/backdoor.md>), [lazarus](<https://devfeed.tech/tags/lazarus.md>), [microsoft](<https://devfeed.tech/tags/microsoft.md>), [research-blog](<https://devfeed.tech/tags/research-blog.md>)

### AI overview

Lazarus can backdoor an LLM by reprogramming fewer than 10 dormant neurons, without training. A hidden trigger can make an otherwise normal agent emit a command that runs with the agent's granted access, even when the model passes standard evaluations. The technique requires write access to model weights, so the excerpt says internally developed models are exposed too. Frontier agents may detect anomalous edits, but scanning cannot prove a model is safe; reliable detection requires comparison with original weights, making provenance and weight integrity central. Microsoft Foundry ties open-weight models to trusted publishers and scans their container environment. Microsoft is developing weight scanning and red-teaming tools and planning signed models with verifiable provenance and tamper-evident audits. The excerpt also recommends limiting agent permissions and monitoring their actions.

### Source excerpt

Microsoft researchers have discovered "Lazarus," a backdoor technique that can reprogram dormant neurons in an AI model, causing an otherwise normal AI agent to execute malicious commands when a hidden trigger appears. The threat is especially serious for AI agents with tool or terminal access. A triggered model could quietly download and run attacker code, [...] The post Lazare, veni foras: Reprogramming Dormant Neurons to Backdoor AI Agents appeared first on Microsoft Research.

## Formally verifying agent-written full-stack web apps with PDD Kit

DevFeed: [Formally verifying agent-written full-stack web apps with PDD Kit](<https://devfeed.tech/articles/formally-verifying-agent-written-full-stack-web-apps-with-pdd-kit-120876.md>)

Original publisher: [Read original article](<https://www.microsoft.com/en-us/research/blog/formally-verifying-agent-written-full-stack-web-apps-with-pdd-kit/>)

Author: Alex Lavaee, Matthai Philipose

Published: 2026-11-05T16:00:00Z

Content type: article

Language: en

Sources: [Microsoft Research](<https://devfeed.tech/sources/microsoft-research.md>)

Topics: [Spec Driven Development](<https://devfeed.tech/topics/spec-driven-development.md>), [Full Stack Development](<https://devfeed.tech/topics/full-stack-development.md>), [CI/CD](<https://devfeed.tech/topics/cicd.md>), [Lean](<https://devfeed.tech/topics/lean.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [apps](<https://devfeed.tech/tags/apps.md>), [code](<https://devfeed.tech/tags/code.md>), [developer](<https://devfeed.tech/tags/developer.md>), [developers](<https://devfeed.tech/tags/developers.md>), [formal](<https://devfeed.tech/tags/formal.md>), [full-stack](<https://devfeed.tech/tags/full-stack.md>), [generate](<https://devfeed.tech/tags/generate.md>), [helps](<https://devfeed.tech/tags/helps.md>), [input](<https://devfeed.tech/tags/input.md>), [microsoft](<https://devfeed.tech/tags/microsoft.md>), [microsoft-research](<https://devfeed.tech/tags/microsoft-research.md>), [post](<https://devfeed.tech/tags/post.md>), [practical](<https://devfeed.tech/tags/practical.md>), [proofs](<https://devfeed.tech/tags/proofs.md>), [research-blog](<https://devfeed.tech/tags/research-blog.md>), [web](<https://devfeed.tech/tags/web.md>), [web-apps](<https://devfeed.tech/tags/web-apps.md>)

### AI overview

PDD Kit turns a developer's plain-language requirements into a reviewable intent artifact and Verus specifications, then uses coding agents to implement and prove specified properties for every input. In a Rust Pet Store login example, the proof covers account selection, password gating, token claims, and refusal outcomes; tests cover behavior outside the proof. The developer reviews intent and explicitly accepts assumptions at the trusted boundary, including database lookups, bcrypt, clock access, token signing, and request handling. Those assumptions remain limits: the proof cannot establish that the specification matches intent or that trusted components behave as assumed, and it says nothing about completion, timing, or unspecified requirements. PDD Kit is a research prototype that currently verifies Rust; its workflow adds proof checks to CI alongside tests.

### Source excerpt

AI agents can generate code at a scale that's hard for humans to review line-by-line. PDD Kit helps ensure the code is doing what the developer intended by creating formal specifications and testing against them. It converts developer intent into a Verus specification, proves AI-generated code meets it for every input, and uses tests where proofs aren't practical. Developers review intent, not every line. The post Formally verifying agent-written full-stack web apps with PDD Kit appeared first on Microsoft Research.

## Learn the rules: The hacker mindset

DevFeed: [Learn the rules: The hacker mindset](<https://devfeed.tech/articles/learn-the-rules-the-hacker-mindset-116091.md>)

Original publisher: [Read original article](<https://www.microsoft.com/en-us/research/podcast/learn-the-rules-the-hacker-mindset/>)

Author: Weishung Liu, Mike Walker

Published: 2026-11-03T17:14:38Z

Content type: article

Language: en

Sources: [Microsoft Research](<https://devfeed.tech/sources/microsoft-research.md>)

Topics: [Cybersecurity](<https://devfeed.tech/topics/cybersecurity.md>), [ctf](<https://devfeed.tech/topics/ctf.md>), [engineering-leadership](<https://devfeed.tech/topics/engineering-leadership.md>), [trust](<https://devfeed.tech/topics/trust.md>), [Sports](<https://devfeed.tech/topics/sports.md>), [Tool](<https://devfeed.tech/topics/tool.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>)

Tags: [about](<https://devfeed.tech/tags/about.md>), [ai](<https://devfeed.tech/tags/ai.md>), [career](<https://devfeed.tech/tags/career.md>), [culture](<https://devfeed.tech/tags/culture.md>), [curiosity](<https://devfeed.tech/tags/curiosity.md>), [cybersecurity](<https://devfeed.tech/tags/cybersecurity.md>), [first](<https://devfeed.tech/tags/first.md>), [hacker](<https://devfeed.tech/tags/hacker.md>), [hacker-culture](<https://devfeed.tech/tags/hacker-culture.md>), [leadership](<https://devfeed.tech/tags/leadership.md>), [learn](<https://devfeed.tech/tags/learn.md>), [manager](<https://devfeed.tech/tags/manager.md>), [microsoft](<https://devfeed.tech/tags/microsoft.md>), [microsoft-research](<https://devfeed.tech/tags/microsoft-research.md>), [microsoft-research-podcast](<https://devfeed.tech/tags/microsoft-research-podcast.md>), [mindset](<https://devfeed.tech/tags/mindset.md>), [post](<https://devfeed.tech/tags/post.md>), [problems](<https://devfeed.tech/tags/problems.md>), [requires](<https://devfeed.tech/tags/requires.md>), [research](<https://devfeed.tech/tags/research.md>), [rules](<https://devfeed.tech/tags/rules.md>)

### AI overview

Mike Walker argues that effective cybersecurity depends on understanding how rules operate in social contexts: each security demand asks users for trust, and squandered trust weakens the security regime. He describes how competitive Capture the Flag hacking informed DARPA's Cyber Grand Challenge, an autonomous tournament in which machines found and patched vulnerabilities; he characterizes its 2016 systems as expert systems, with their tool stacks serving as harnesses for later model-powered systems. Walker says offensive and defensive capabilities are often dual use, making red and blue team divisions incomplete. His team builds defensive tools with awareness that they can be targeted. Project Freta aims to counter tampering by analyzing memory signals acquired from cloud hardware rather than software that attackers could remove or alter. He also reflects that technical teams need coordination and shared awareness, not just individual skill, and that leaders may make decisions based on information others lack.

### Source excerpt

Research Manager Mike Walker reflects on a career spanning hacker culture, cybersecurity, and AI, sharing lessons about trust, teamwork, leadership, and why solving difficult problems requires curiosity, judgment, agility--and an understanding of the rules. The post Learn the rules: The hacker mindset appeared first on Microsoft Research.

## From synthetic enterprise data evaluations to real Microsoft Copilot AI confidence

DevFeed: [From synthetic enterprise data evaluations to real Microsoft Copilot AI confidence](<https://devfeed.tech/articles/from-synthetic-enterprise-data-evaluations-to-real-microsoft-copilot-ai-confidence-109787.md>)

Original publisher: [Read original article](<https://www.microsoft.com/en-us/research/blog/from-synthetic-enterprise-data-evaluations-to-real-microsoft-copilot-ai-confidence/>)

Author: Somya Chatterjee, Harshita Sahijwani, Min Chen, Sushmitha Rajeswari Muppa, Dayquan Julienne, Irene Shaffer, Chandra Inguva, Jenny Chen, Minha Hwang, Sam Ade Jacobs

Published: 2026-10-30T13:05:28Z

Content type: article

Language: en

Sources: [Microsoft Research](<https://devfeed.tech/sources/microsoft-research.md>)

Topics: [synthetic-data](<https://devfeed.tech/topics/synthetic-data.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [Data pipelines](<https://devfeed.tech/topics/data-pipelines.md>), [managing large-scale text data in AI](<https://devfeed.tech/topics/managing-large-scale-text-data-in-ai.md>), [enterprise data protection](<https://devfeed.tech/topics/enterprise-data-protection.md>), [trust](<https://devfeed.tech/topics/trust.md>), [data-governance](<https://devfeed.tech/topics/data-governance.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [complexity](<https://devfeed.tech/tags/complexity.md>), [confidence](<https://devfeed.tech/tags/confidence.md>), [copilot](<https://devfeed.tech/tags/copilot.md>), [data](<https://devfeed.tech/tags/data.md>), [enterprise](<https://devfeed.tech/tags/enterprise.md>), [enterprise-data](<https://devfeed.tech/tags/enterprise-data.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [evaluations](<https://devfeed.tech/tags/evaluations.md>), [first](<https://devfeed.tech/tags/first.md>), [full](<https://devfeed.tech/tags/full.md>), [information](<https://devfeed.tech/tags/information.md>), [microsoft](<https://devfeed.tech/tags/microsoft.md>), [microsoft-copilot](<https://devfeed.tech/tags/microsoft-copilot.md>), [organizations](<https://devfeed.tech/tags/organizations.md>), [post](<https://devfeed.tech/tags/post.md>), [requires](<https://devfeed.tech/tags/requires.md>), [research-blog](<https://devfeed.tech/tags/research-blog.md>), [scale](<https://devfeed.tech/tags/scale.md>), [synthetic](<https://devfeed.tech/tags/synthetic.md>), [test](<https://devfeed.tech/tags/test.md>)

### AI overview

Microsoft describes an Industry Evaluation Data Generation Platform for testing Copilot with synthetic enterprise tenants rather than customer data. Its environments model evolving organizations and workflows across pharmaceutical, capital markets, and IT/software industries, using public O*NET data, job postings, and SEC filings to guide realism. A six-stage automated pipeline adapts workplace tasks into industry-specific queries grounded in those environments, then checks accuracy and uniqueness. Internal quality measures cover structural fidelity, contextual relevance, behavioral realism, and content safety. Microsoft reports a 39% improvement in its composite synthetic-data quality score, creation time reduced from multiple days to automated runs, and estimated savings of about $100,000 per dataset and $300,000 overall. The text says the datasets use no customer data and are screened for harmful or identifying content; public sources receive legal and policy review.

### Source excerpt

Evaluating Microsoft Copilot at enterprise scale requires test data that captures the full complexity of real organizations, yet real enterprise data contains sensitive information that cannot be used for evaluation under any circumstances. The post From synthetic enterprise data evaluations to real Microsoft Copilot AI confidence appeared first on Microsoft Research.

## Flint-interactive: A language for creating interactive data visualizations

DevFeed: [Flint-interactive: A language for creating interactive data visualizations](<https://devfeed.tech/articles/flint-interactive-a-language-for-creating-interactive-data-visualizations-96041.md>)

Original publisher: [Read original article](<https://www.microsoft.com/en-us/research/blog/flint-interactive-a-language-for-creating-interactive-data-visualizations/>)

Author: Chenglong Wang, Yuwei Xiao, Alper Sarikaya, Michel Galley, Chris Quirk, Jianfeng Gao

Published: 2026-10-22T16:00:00Z

Content type: article

Language: en

Sources: [Microsoft Research](<https://devfeed.tech/sources/microsoft-research.md>)

Topics: [AI for interactive applications](<https://devfeed.tech/topics/ai-for-interactive-applications.md>), [AI Chat](<https://devfeed.tech/topics/ai-chat.md>), [data](<https://devfeed.tech/topics/data.md>), [interactive web design](<https://devfeed.tech/topics/interactive-web-design.md>)

Tags: [2](<https://devfeed.tech/tags/2.md>), [3](<https://devfeed.tech/tags/3.md>), [actions](<https://devfeed.tech/tags/actions.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [allows](<https://devfeed.tech/tags/allows.md>), [also](<https://devfeed.tech/tags/also.md>), [analyze](<https://devfeed.tech/tags/analyze.md>), [annotation](<https://devfeed.tech/tags/annotation.md>), [api](<https://devfeed.tech/tags/api.md>), [app](<https://devfeed.tech/tags/app.md>), [data](<https://devfeed.tech/tags/data.md>), [explore](<https://devfeed.tech/tags/explore.md>), [filter](<https://devfeed.tech/tags/filter.md>), [flint](<https://devfeed.tech/tags/flint.md>), [interactive](<https://devfeed.tech/tags/interactive.md>), [language](<https://devfeed.tech/tags/language.md>), [research-blog](<https://devfeed.tech/tags/research-blog.md>), [view](<https://devfeed.tech/tags/view.md>)

### AI overview

Flint-interactive lets agents and developers specify chart interactions at a semantic level, while Flint's compiler translates user actions into events and update operators into rendered changes. This avoids requiring renderer-specific code and is intended to reduce interaction logic, layout, and debugging effort. In chat, Flint-MCP and MCP Apps can generate an interaction_spec alongside a Flint chart spec. In applications, Flint Canvas can emit semantic events that handlers use to filter tables, update charts or widgets, or change app state. Bespoke behaviors can be defined with triggers and custom handlers. The excerpt describes support for clicking, filtering, highlighting, panning, zooming, and connected views; it does not provide benchmarks for cost or reliability.

### Source excerpt

Flint-interactive gives AI agents a language for creating interactive visualizations. Click a chart, filter data, highlight related content, and explore beyond the view the author originally chose. The post Flint-interactive: A language for creating interactive data visualizations appeared first on Microsoft Research.

## Solving a 31-year-old grand challenge in mathematics - the KLS conjecture

DevFeed: [Solving a 31-year-old grand challenge in mathematics - the KLS conjecture](<https://devfeed.tech/articles/solving-a-31-year-old-grand-challenge-in-mathematics-the-kls-conjecture-92996.md>)

Original publisher: [Read original article](<https://www.microsoft.com/en-us/research/blog/solving-a-31-year-old-grand-challenge-in-mathematics-the-kls-conjecture/>)

Author: Xinzhi Zhang

Published: 2026-10-21T16:00:00Z

Content type: news

Language: en

Sources: [Microsoft Research](<https://devfeed.tech/sources/microsoft-research.md>)

Topics: [Mathematics](<https://devfeed.tech/topics/mathematics.md>), [Computer science](<https://devfeed.tech/topics/computer-science.md>), [Parallel reasoning](<https://devfeed.tech/topics/parallel-reasoning.md>), [Automated reasoning](<https://devfeed.tech/topics/automated-reasoning.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [agent](<https://devfeed.tech/tags/agent.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [ai-collaboration](<https://devfeed.tech/tags/ai-collaboration.md>), [ai-models](<https://devfeed.tech/tags/ai-models.md>), [ai-systems](<https://devfeed.tech/tags/ai-systems.md>), [ai-tools](<https://devfeed.tech/tags/ai-tools.md>), [algorithms](<https://devfeed.tech/tags/algorithms.md>), [breakthrough](<https://devfeed.tech/tags/breakthrough.md>), [central](<https://devfeed.tech/tags/central.md>), [computing](<https://devfeed.tech/tags/computing.md>), [efficiency](<https://devfeed.tech/tags/efficiency.md>), [mathematics](<https://devfeed.tech/tags/mathematics.md>), [research](<https://devfeed.tech/tags/research.md>), [research-blog](<https://devfeed.tech/tags/research-blog.md>), [science](<https://devfeed.tech/tags/science.md>), [spaces](<https://devfeed.tech/tags/spaces.md>), [theoretical](<https://devfeed.tech/tags/theoretical.md>)

### AI overview

The authors report proving the Kannan-Lovász-Simonovits conjecture, showing that for log-concave distributions the best flat cut captures the narrowest bottleneck within a universal constant, independent of dimension. Their proof repeatedly applies a recursive refinement and shows that the errors remain bounded. AI agents explored more than 100 candidate approaches and helped with derivations and checks; researchers selected the promising ideas and built and verified the proof. Unguided attempts with leading models, using roughly the same total token budget, did not solve the problem. The text reports parallel independent proofs and a Lean 4 formalization as additional validation, and says the authors plan to release a streamlined proof after final review.

### Source excerpt

For decades, a fundamental question about the difficulty of computing in high-dimensional spaces has resisted proof. New research resolves this question, removing a major theoretical obstacle to understanding the efficiency of randomized algorithms used across science and computing. The Kannan-Lovász-Simonovits (KLS) conjecture was proposed in 1995. It asks whether, for a broad class of high-dimensional [...] The post Solving a 31-year-old grand challenge in mathematics - the KLS conjecture appeared first on Microsoft Research.

## Called to serve: Tech, research, and positive impact

DevFeed: [Called to serve: Tech, research, and positive impact](<https://devfeed.tech/articles/called-to-serve-tech-research-and-positive-impact-91275.md>)

Original publisher: [Read original article](<https://www.microsoft.com/en-us/research/podcast/called-to-serve-tech-research-and-positive-impact-with-chris-white/>)

Author: Weishung Liu, Chris White

Published: 2026-10-20T16:00:00Z

Content type: article

Language: en

Sources: [Microsoft Research](<https://devfeed.tech/sources/microsoft-research.md>)

Topics: [Tool](<https://devfeed.tech/topics/tool.md>), [Data Mining & Modeling](<https://devfeed.tech/topics/data-mining-modeling.md>), [Sports](<https://devfeed.tech/topics/sports.md>), [Image](<https://devfeed.tech/topics/image.md>)

Tags: [academic](<https://devfeed.tech/tags/academic.md>), [academic-research](<https://devfeed.tech/tags/academic-research.md>), [advanced](<https://devfeed.tech/tags/advanced.md>), [advice](<https://devfeed.tech/tags/advice.md>), [afghanistan](<https://devfeed.tech/tags/afghanistan.md>), [data](<https://devfeed.tech/tags/data.md>), [data-analysis](<https://devfeed.tech/tags/data-analysis.md>), [impact](<https://devfeed.tech/tags/impact.md>), [lab](<https://devfeed.tech/tags/lab.md>), [manager](<https://devfeed.tech/tags/manager.md>), [microsoft](<https://devfeed.tech/tags/microsoft.md>), [microsoft-research](<https://devfeed.tech/tags/microsoft-research.md>), [microsoft-research-podcast](<https://devfeed.tech/tags/microsoft-research-podcast.md>), [real-world](<https://devfeed.tech/tags/real-world.md>), [research](<https://devfeed.tech/tags/research.md>), [serve](<https://devfeed.tech/tags/serve.md>), [talks](<https://devfeed.tech/tags/talks.md>), [tech](<https://devfeed.tech/tags/tech.md>), [tools](<https://devfeed.tech/tags/tools.md>), [work](<https://devfeed.tech/tags/work.md>), [world](<https://devfeed.tech/tags/world.md>)

### AI overview

Chris White describes how his work shifted from academic research to applying technology to urgent public problems. In a 2010 DARPA effort, a multidisciplinary team explored whether data analysis could help prevent violence and protect troops in Afghanistan; White says he was sent to test the work in the field within a week, initially expecting a nine-day deployment that extended to 90 days. Later, as a DARPA program manager, he developed Memex tools to connect information across less-visible parts of the web. He says this evidence helped trafficking investigators in New York City increase prosecutions tenfold. White links his choices to service, humility, and selecting problems whose outcomes matter. At Microsoft Research's Catalyst Lab, he values research freedom paired with ways to test value and usability, and cites cyber defense and content analysis as current opportunities. He also reflects that slowing down, especially when making decisions about people, helps avoid problems and gain perspective.

### Source excerpt

Lab Director Chris White has worked on research challenges with real-world implications--from new approaches to wartime data analysis to tools for combating human trafficking. He talks to program manager Weishung Liu about the influences that led to the work and more. The post Called to serve: Tech, research, and positive impact appeared first on Microsoft Research.

## Proof-based general-purpose computing with Magnetar

DevFeed: [Proof-based general-purpose computing with Magnetar](<https://devfeed.tech/articles/proof-based-general-purpose-computing-with-magnetar-88214.md>)

Original publisher: [Read original article](<https://www.microsoft.com/en-us/research/blog/proof-based-general-purpose-computing-with-magnetar/>)

Author: Srinath Setty

Published: 2026-10-15T16:00:00Z

Content type: article

Language: en

Sources: [Microsoft Research](<https://devfeed.tech/sources/microsoft-research.md>)

Topics: [Zero-knowledge proof](<https://devfeed.tech/topics/zkp.md>), [virtualization](<https://devfeed.tech/topics/virtualization.md>)

Tags: [computing](<https://devfeed.tech/tags/computing.md>), [proofs](<https://devfeed.tech/tags/proofs.md>), [research-blog](<https://devfeed.tech/tags/research-blog.md>), [security](<https://devfeed.tech/tags/security.md>)

### AI overview

Microsoft Research presents Magnetar, a zero-knowledge virtual machine for proving ordinary program executions. It supports Rust, Go, and C programs compiled for RISC-V, and the article reports faster proving than comparable systems alongside security based on standard hash functions. Proofs can be checked without rerunning the workload or revealing private inputs.

### Source excerpt

Zero-knowledge proofs can verify computing results without rerunning workloads or exposing sensitive data. Magnetar helps bring them into the mainstream with support for standard programming languages, major speed gains, and quantum-resistant security. The post Proof-based general-purpose computing with Magnetar appeared first on Microsoft Research.

## Simulating student learning for better AI tutoring

DevFeed: [Simulating student learning for better AI tutoring](<https://devfeed.tech/articles/simulating-student-learning-for-better-ai-tutoring-85614.md>)

Original publisher: [Read original article](<https://www.microsoft.com/en-us/research/blog/simulating-student-learning-for-better-ai-tutoring/>)

Author: Ke Yang, Chenglong Wang, Michel Galley, Jeevana Priya Inala, ChengXiang Zhai, Jianfeng Gao

Published: 2026-10-12T16:00:00Z

Content type: article

Language: en

Sources: [Microsoft Research](<https://devfeed.tech/sources/microsoft-research.md>)

Topics: [AI Development](<https://devfeed.tech/topics/ai-development.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [research](<https://devfeed.tech/tags/research.md>), [research-blog](<https://devfeed.tech/tags/research-blog.md>)

### AI overview

StudentSim is a research system that creates personalized simulators of learners to help researchers evaluate and improve AI tutors before testing them with real students. It trains on pooled and student-specific de-identified records, then assesses whether simulators reflect individual behavior and respond to tutor guidance. Across chess, second-language English writing, and mathematics, it outperforms leading baselines. In a chess reinforcement learning study, tutors trained with StudentSim-generated feedback received higher expert ratings and achieved 90.5% accuracy, compared with 75.7% for the no-reinforcement-learning baseline and 71.6% for the GPT-5.4 reward condition. The work is a proof of concept; real learner studies remain necessary.

### Source excerpt

StudentSim is a research system for evaluating and improving AI tutoring tools before they are tested with learners. It models how individuals may engage with tutors, giving feedback to improve tutor performance across subjects like chess, writing, and mathematics. The post Simulating student learning for better AI tutoring appeared first on Microsoft Research.

## Agent Lightning v1.0: A 3,500-Line Lightweight Agentic RL Framework for Training Agents with Real Harnesses

DevFeed: [Agent Lightning v1.0: A 3,500-Line Lightweight Agentic RL Framework for Training Agents with Real Harnesses](<https://devfeed.tech/articles/agent-lightning-v1-0-a-3-500-line-lightweight-agentic-rl-framework-for-training-agents-with-real-harnesses-66970.md>)

Original publisher: [Read original article](<https://www.microsoft.com/en-us/research/blog/agent-lightning-v1-0-a-3500-line-lightweight-agentic-rl-framework-for-training-agents-with-real-harnesses/>)

Author: Zhiyuan He, Yuqing Yang

Published: 2026-10-07T16:00:00Z

Content type: article

Language: en

Sources: [Microsoft Research](<https://devfeed.tech/sources/microsoft-research.md>)

Topics: [grpo](<https://devfeed.tech/topics/grpo.md>), [agentic-coding](<https://devfeed.tech/topics/agentic-coding.md>), [Agent Harness](<https://devfeed.tech/topics/agent-harness.md>), [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>)

Tags: [agent-harness](<https://devfeed.tech/tags/agent-harness.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [api](<https://devfeed.tech/tags/api.md>), [api-gateway](<https://devfeed.tech/tags/api-gateway.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [research-blog](<https://devfeed.tech/tags/research-blog.md>), [rl](<https://devfeed.tech/tags/rl.md>)

### AI overview

Microsoft Research Asia presents Agent Lightning v1.0, an open-source framework that connects existing agent harnesses to reinforcement learning without rebuilding their interaction loops. The framework includes an API gateway, rollout controller, and trainer, supports running agents as Kubernetes jobs, and reports a roughly twofold speedup for collocated asynchronous RL. In a coding-agent experiment, training Qwen3.5-9B on about 6,000 samples raised SWE-bench Verified Pass@1 from 41.8% to 56.4%.

### Source excerpt

Training AI agents with reinforcement learning can be challenging because their tools, context, and decision-making are managed by complex frameworks. Agent Lightning connects existing agents to RL training, making it easier to improve them without rebuilding them. The post Agent Lightning v1.0: A 3,500-Line Lightweight Agentic RL Framework for Training Agents with Real Harnesses appeared first on Microsoft Research.

## What AI gets wrong and what failure teaches us

DevFeed: [What AI gets wrong and what failure teaches us](<https://devfeed.tech/articles/what-ai-gets-wrong-and-what-failure-teaches-us-65885.md>)

Original publisher: [Read original article](<https://www.microsoft.com/en-us/research/podcast/what-ai-gets-wrong-and-what-failure-teaches-us/>)

Author: Chad Atalla, Jennifer Neville

Published: 2026-10-06T16:19:06Z

Content type: opinion

Language: en

Sources: [Microsoft Research](<https://devfeed.tech/sources/microsoft-research.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Hallucination detection](<https://devfeed.tech/topics/hallucination-detection.md>), [AI-enhanced product development](<https://devfeed.tech/topics/ai-enhanced-product-development.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-evaluation](<https://devfeed.tech/tags/ai-evaluation.md>), [ai-research](<https://devfeed.tech/tags/ai-research.md>), [ai-systems](<https://devfeed.tech/tags/ai-systems.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [microsoft-research-podcast](<https://devfeed.tech/tags/microsoft-research-podcast.md>)

### AI overview

In this Microsoft Research Podcast conversation, AI researcher Jennifer Neville discusses evaluating AI systems in realistic, multi-turn and long-horizon workflows. She describes performance gaps missed by simple benchmarks, including failures that accumulate during document-editing tasks, and explains how examining data and observing real user needs can guide improvements. She also advises users to verify AI outputs and work with current systems under human oversight.

### Source excerpt

Jennifer Neville did not want to go into computer science--but that's exactly where she landed. Neville discusses the starts and stops that led to her professional sweet spot and her work identifying "surprising failures" making it hard for AI to handle complexity. The post What AI gets wrong and what failure teaches us appeared first on Microsoft Research.

## Forecasting space weather risks on power grids

DevFeed: [Forecasting space weather risks on power grids](<https://devfeed.tech/articles/forecasting-space-weather-risks-on-power-grids-62874.md>)

Original publisher: [Read original article](<https://www.microsoft.com/en-us/research/blog/forecasting-space-weather-risks-on-power-grids/>)

Author: Rohan Kannan

Published: 2026-09-30T16:00:00Z

Content type: article

Language: en

Sources: [Microsoft Research](<https://devfeed.tech/sources/microsoft-research.md>)

Topics: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Climate / Weather / Ocean Modeling](<https://devfeed.tech/topics/climate-weather-ocean-modeling.md>), [clean energy data centers](<https://devfeed.tech/topics/clean-energy-data-centers.md>)

Tags: [critical-infrastructure](<https://devfeed.tech/tags/critical-infrastructure.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [forecasting](<https://devfeed.tech/tags/forecasting.md>), [gps](<https://devfeed.tech/tags/gps.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [power](<https://devfeed.tech/tags/power.md>), [power-grids](<https://devfeed.tech/tags/power-grids.md>), [research-blog](<https://devfeed.tech/tags/research-blog.md>), [space](<https://devfeed.tech/tags/space.md>)

### AI overview

A machine learning pipeline estimates space-weather risks for 66,935 substations in the continental United States. It combines solar-wind forecasts, magnetic indices, local geology, and grid data to provide location-specific estimates 30-60 minutes ahead. During the evaluation period, it detected nearly 80% of major space-weather events.

### Source excerpt

Extreme space-weather events can damage power systems on Earth and degrade GPS accuracy and satellite operations. A new machine learning system can predict where damage is likely to occur 30-60 minutes before a storm arrives. The post Forecasting space weather risks on power grids appeared first on Microsoft Research.

## Introducing Quine: An AI research system designed for the complexity of biology

DevFeed: [Introducing Quine: An AI research system designed for the complexity of biology](<https://devfeed.tech/articles/introducing-quine-an-ai-research-system-designed-for-the-complexity-of-biology-61639.md>)

Original publisher: [Read original article](<https://www.microsoft.com/en-us/research/blog/introducing-quine-an-ai-research-system-designed-for-the-complexity-of-biology/>)

Author: Nicolo Fusi, Jonathan M. Carlson

Published: 2026-09-29T14:00:02Z

Content type: article

Language: en

Sources: [Microsoft Research](<https://devfeed.tech/sources/microsoft-research.md>)

Topics: [World models](<https://devfeed.tech/topics/world-models.md>), [AI Research](<https://devfeed.tech/topics/ai-research.md>), [AI for science](<https://devfeed.tech/topics/ai-for-science.md>), [AI-enhanced product development](<https://devfeed.tech/topics/ai-enhanced-product-development.md>), [human review](<https://devfeed.tech/topics/human-review.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-research](<https://devfeed.tech/tags/ai-research.md>), [biomedical](<https://devfeed.tech/tags/biomedical.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [detection](<https://devfeed.tech/tags/detection.md>), [discovery](<https://devfeed.tech/tags/discovery.md>), [drive](<https://devfeed.tech/tags/drive.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [genomics](<https://devfeed.tech/tags/genomics.md>), [imaging](<https://devfeed.tech/tags/imaging.md>), [mit](<https://devfeed.tech/tags/mit.md>), [models](<https://devfeed.tech/tags/models.md>), [next](<https://devfeed.tech/tags/next.md>), [proteins](<https://devfeed.tech/tags/proteins.md>), [research](<https://devfeed.tech/tags/research.md>), [research-blog](<https://devfeed.tech/tags/research-blog.md>), [techniques](<https://devfeed.tech/tags/techniques.md>), [technology](<https://devfeed.tech/tags/technology.md>)

### AI overview

Quine is an experimental Microsoft Research effort to build a multimodal world model of biology and connect it with scientific tools, literature, researchers, and lab experiments. In collaboration with Broad Institute researchers, the system helped prioritize compounds predicted to drive therapeutic tumor-state shifts, with several leading candidates validated in wet-lab assays. The system is for research use, and its outputs require qualified review and scientific validation.

### Source excerpt

Biology doesn't operate in silos, and neither should the AI representation of it. Quine is an early-stage research effort to create a multimodal world model of biology. By connecting insights across biological scales and modalities, Quine helps scientists computationally search a space far larger than intuition allows and prioritize hypotheses before they reach the lab. Experimental results provide important feedback, helping researchers sharpen future research directions. The post Introducing Quine: An AI research system designed for the complexity of biology appeared first on Microsoft Research.

## One year in: How Microsoft Research Asia - Singapore is advancing research, partnership and talent for real-world impact

DevFeed: [One year in: How Microsoft Research Asia - Singapore is advancing research, partnership and talent for real-world impact](<https://devfeed.tech/articles/one-year-in-how-microsoft-research-asia-singapore-is-advancing-research-partnership-and-talent-for-real-world-impact-61376.md>)

Original publisher: [Read original article](<https://www.microsoft.com/en-us/research/blog/one-year-in-how-microsoft-research-asia-singapore-is-advancing-research-partnership-and-talent-for-real-world-impact/>)

Author: Microsoft Research Team

Published: 2026-09-28T21:00:00Z

Content type: article

Language: en

Sources: [Microsoft Research](<https://devfeed.tech/sources/microsoft-research.md>)

Topics: [Frontier AI](<https://devfeed.tech/topics/frontier-ai.md>), [AI-enhanced product development](<https://devfeed.tech/topics/ai-enhanced-product-development.md>), [AI interaction best practices](<https://devfeed.tech/topics/ai-interaction-best-practices.md>), [Product Management](<https://devfeed.tech/topics/product-management.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [frontier-ai](<https://devfeed.tech/tags/frontier-ai.md>), [microsoft](<https://devfeed.tech/tags/microsoft.md>), [partnership](<https://devfeed.tech/tags/partnership.md>), [research-blog](<https://devfeed.tech/tags/research-blog.md>), [singapore](<https://devfeed.tech/tags/singapore.md>)

### AI overview

Microsoft Research Asia's Singapore lab describes its first year building partnerships and expanding frontier AI research aimed at real-world use. Its work includes agentic systems, AI-native research practices, and collaborations applying AI in healthcare and other domains.

### Source excerpt

Since launching a year ago, the Microsoft Research Asia -- Singapore lab has established a strong foundation, deepened collaboration across government, academia, and industry, and explored how frontier AI research can create real-world value. The post One year in: How Microsoft Research Asia - Singapore is advancing research, partnership and talent for real-world impact appeared first on Microsoft Research.

## Offloaded inference for real-world physical AI robotics

DevFeed: [Offloaded inference for real-world physical AI robotics](<https://devfeed.tech/articles/offloaded-inference-for-real-world-physical-ai-robotics-58925.md>)

Original publisher: [Read original article](<https://www.microsoft.com/en-us/research/blog/offloaded-inference-for-real-world-physical-ai-robotics/>)

Author: Ganesh Ananthanarayanan, Matthew Balkwill, Xenofon Foukas, Sanjeev Mehrotra, Bozidar Radunovic, Connor Settle, Ankit Verma, David White, Shawn Cicoria, Mark Martin, Rachel Johnson, Mayur Patel

Published: 2026-09-23T16:01:36Z

Content type: article

Language: en

Sources: [Microsoft Research](<https://devfeed.tech/sources/microsoft-research.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Inference](<https://devfeed.tech/topics/inference.md>), [Physical AI](<https://devfeed.tech/topics/physical-ai.md>), [Robotics](<https://devfeed.tech/topics/robotics.md>), [cloud-infrastructure](<https://devfeed.tech/topics/cloud-infrastructure.md>), [Kubernetes](<https://devfeed.tech/topics/kubernetes.md>), [toolchain](<https://devfeed.tech/topics/toolchain.md>), [Scalability](<https://devfeed.tech/topics/scalability.md>), [Cloud](<https://devfeed.tech/topics/cloud.md>), [GPU](<https://devfeed.tech/topics/gpu.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [ai-inference](<https://devfeed.tech/tags/ai-inference.md>), [ai-robotics](<https://devfeed.tech/tags/ai-robotics.md>), [cloud](<https://devfeed.tech/tags/cloud.md>), [distributed](<https://devfeed.tech/tags/distributed.md>), [edge](<https://devfeed.tech/tags/edge.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [inference](<https://devfeed.tech/tags/inference.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [kubernetes](<https://devfeed.tech/tags/kubernetes.md>), [physical-ai](<https://devfeed.tech/tags/physical-ai.md>), [research](<https://devfeed.tech/tags/research.md>), [research-blog](<https://devfeed.tech/tags/research-blog.md>), [robotics](<https://devfeed.tech/tags/robotics.md>)

### AI overview

Microsoft Research presents findings on offloading physical AI inference from robot-mounted GPUs to edge or cloud GPUs. The approach improved task success, enabled larger models, and extended robot operating time in representative mobile manipulation workloads.

### Source excerpt

Robots are getting smarter, but how can their hardware match that growth? New Microsoft Research findings show that moving AI inference beyond the robot can improve task success, boost efficiency, and support more advanced physical AI workloads. The post Offloaded inference for real-world physical AI robotics appeared first on Microsoft Research.

## GigaPath-Flash and GigaTIME-Flash: Toward population-scale discovery with efficient pathology foundation models

DevFeed: [GigaPath-Flash and GigaTIME-Flash: Toward population-scale discovery with efficient pathology foundation models](<https://devfeed.tech/articles/gigapath-flash-and-gigatime-flash-toward-population-scale-discovery-with-efficient-pathology-foundation-models-6796.md>)

Original publisher: [Read original article](<https://www.microsoft.com/en-us/research/blog/gigapath-flash-and-gigatime-flash-toward-population-scale-discovery-with-efficient-pathology-foundation-models/>)

Author: Naoto Usuyama, Jeya Maria Jose Valanarasu, Tristan Naumann

Published: 2026-08-31T16:00:00Z

Content type: article

Language: en

Sources: [Microsoft Research](<https://devfeed.tech/sources/microsoft-research.md>)

Topics: [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>), [dataset](<https://devfeed.tech/topics/dataset.md>), [data](<https://devfeed.tech/topics/data.md>)

Tags: [analysis](<https://devfeed.tech/tags/analysis.md>), [data](<https://devfeed.tech/tags/data.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [foundation-models](<https://devfeed.tech/tags/foundation-models.md>), [models](<https://devfeed.tech/tags/models.md>), [performance](<https://devfeed.tech/tags/performance.md>), [research](<https://devfeed.tech/tags/research.md>), [research-blog](<https://devfeed.tech/tags/research-blog.md>), [scale](<https://devfeed.tech/tags/scale.md>), [validation](<https://devfeed.tech/tags/validation.md>)

### AI overview

GigaPath-Flash and GigaTIME-Flash are efficient pathology foundation models designed to reduce computational demands while maintaining strong performance. They enable repeated analysis of larger cancer cohorts and support population-scale research into disease biology, biomarkers, and clinical outcomes.

### Source excerpt

What if pathology foundation models could do more with less? GigaPath-Flash and GigaTIME-Flash cut computational demands while maintaining strong performance, opening the door to larger studies and broader exploration. The post GigaPath-Flash and GigaTIME-Flash: Toward population-scale discovery with efficient pathology foundation models appeared first on Microsoft Research.

## Broadening access to Skala creates a faster path to predictive DFT

DevFeed: [Broadening access to Skala creates a faster path to predictive DFT](<https://devfeed.tech/articles/broadening-access-to-skala-creates-a-faster-path-to-predictive-dft-6783.md>)

Original publisher: [Read original article](<https://www.microsoft.com/en-us/research/blog/broadening-access-to-skala-creates-a-faster-path-to-predictive-dft/>)

Author: Sebastian Ehlert, Stefano Battaglia, Thijs Vogels, Jan Hermann, Jens Wehner, Giulia Luise, Klaas Giesbertz, Chin-Wei Huang, Aaron Kaplan, Kate Milton, Stephanie Marisa Lanius, Derk Kooi, P. Bernát Sza

Published: 2026-08-20T16:00:00Z

Content type: article

Language: en

Sources: [Microsoft Research](<https://devfeed.tech/sources/microsoft-research.md>)

Topics: [Deep neural networks](<https://devfeed.tech/topics/deep-neural-networks.md>), [Benchmark](<https://devfeed.tech/topics/benchmark.md>), [Microsoft](<https://devfeed.tech/topics/microsoft.md>), [data](<https://devfeed.tech/topics/data.md>), [Software](<https://devfeed.tech/topics/software.md>)

Tags: [benchmark](<https://devfeed.tech/tags/benchmark.md>), [chemistry](<https://devfeed.tech/tags/chemistry.md>), [data](<https://devfeed.tech/tags/data.md>), [deep-learning](<https://devfeed.tech/tags/deep-learning.md>), [integration](<https://devfeed.tech/tags/integration.md>), [microsoft](<https://devfeed.tech/tags/microsoft.md>), [performance](<https://devfeed.tech/tags/performance.md>), [research](<https://devfeed.tech/tags/research.md>), [research-blog](<https://devfeed.tech/tags/research-blog.md>), [software](<https://devfeed.tech/tags/software.md>), [workflows](<https://devfeed.tech/tags/workflows.md>)

### AI overview

Microsoft Research presents Skala 1.1, a deep-learning exchange-correlation functional that improves accuracy for molecular simulations while expanding access through integrations with major electronic-structure software. A living benchmark will track the computational performance of future releases.

### Source excerpt

Skala 1.1, the updated deep-learning exchange-correlation functional from Microsoft Research, provides greater accuracy, expanded accessibility across the computational chemistry ecosystem, and a living benchmark to track computational performance. The post Broadening access to Skala creates a faster path to predictive DFT appeared first on Microsoft Research.

## MindTopo reveals VLMs' spatial reasoning abilities

DevFeed: [MindTopo reveals VLMs' spatial reasoning abilities](<https://devfeed.tech/articles/mindtopo-reveals-vlms-spatial-reasoning-abilities-6802.md>)

Original publisher: [Read original article](<https://www.microsoft.com/en-us/research/blog/mindtopo-reveals-vlms-spatial-reasoning-abilities/>)

Author: Yunfei Ge, Anbang Liu, Qineng Wang, Johnalbert Garnica, Zihan Wang, Reuben Tan, Jianfeng Gao, Ruohan Zhang, Yining Hong, Jiajun Wu, Manling Li

Published: 2026-08-12T16:00:00Z

Content type: article

Language: en

Sources: [Microsoft Research](<https://devfeed.tech/sources/microsoft-research.md>)

Topics: [Benchmark](<https://devfeed.tech/topics/benchmark.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [multimodal-ai](<https://devfeed.tech/topics/multimodal-ai.md>), [Language models](<https://devfeed.tech/topics/language-models.md>), [Robotics](<https://devfeed.tech/topics/robotics.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [large-language-models](<https://devfeed.tech/tags/large-language-models.md>), [multimodal](<https://devfeed.tech/tags/multimodal.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [research](<https://devfeed.tech/tags/research.md>), [research-blog](<https://devfeed.tech/tags/research-blog.md>), [testing](<https://devfeed.tech/tags/testing.md>), [vlms](<https://devfeed.tech/tags/vlms.md>)

### AI overview

MindTopo is a benchmark for evaluating whether multimodal large language models can understand and manipulate topological relationships such as connectivity, enclosure, order, separation, and knots. It compares static recognition with interactive planning and finds that current models often lose track of structural relationships during sequences of actions.

### Source excerpt

A path, a fence, a knot. MindTopo sets a new benchmark for testing how AI understands topological relationships and highlights new opportunities to strengthen spatial reasoning and planning. The post MindTopo reveals VLMs' spatial reasoning abilities appeared first on Microsoft Research.

## Introducing CARE-X: Towards Clinically Useful Radiology VLMs with Auxiliary Supervision, Reward-Aligned Learning, and Tool-Augmented Measurement

DevFeed: [Introducing CARE-X: Towards Clinically Useful Radiology VLMs with Auxiliary Supervision, Reward-Aligned Learning, and Tool-Augmented Measurement](<https://devfeed.tech/articles/introducing-care-x-towards-clinically-useful-radiology-vlms-with-auxiliary-supervision-reward-aligned-learning-and-tool-augmented-measurement-6800.md>)

Original publisher: [Read original article](<https://www.microsoft.com/en-us/research/blog/introducing-care-x-towards-clinically-useful-radiology-vlms-with-auxiliary-supervision-reward-aligned-learning-and-tool-augmented-measurement/>)

Author: Mercy Ranjit, Nikhilesh E, Dr. Abhyuday Kumara Swamy, Tanuja Ganu

Published: 2026-08-11T16:00:00Z

Content type: article

Language: en

Sources: [Microsoft Research](<https://devfeed.tech/sources/microsoft-research.md>)

Topics: [vlm](<https://devfeed.tech/topics/vlm.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>), [data](<https://devfeed.tech/topics/data.md>)

Tags: [ai](<https://devfeed.tech/tags/ai.md>), [data](<https://devfeed.tech/tags/data.md>), [generation](<https://devfeed.tech/tags/generation.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [reinforcement-learning](<https://devfeed.tech/tags/reinforcement-learning.md>), [research](<https://devfeed.tech/tags/research.md>), [research-blog](<https://devfeed.tech/tags/research-blog.md>), [tools](<https://devfeed.tech/tags/tools.md>), [vlms](<https://devfeed.tech/tags/vlms.md>)

### AI overview

CARE-X is a research chest X-ray vision-language model that combines free-text report generation, structured diagnostic prediction, and reinforcement learning for multi-task clinical interpretation. The article also describes a separate experiment using deterministic measurement tools with Qwen3-VL-4B-Instruct and reports validation on real-world Indian clinical data, while emphasizing that CARE-X is not approved for clinical use.

### Source excerpt

Radiology AI is evolving beyond report generation. CARE-X explores a unified approach that combines flexible reasoning, calibrated predictions, and measurement-based tools for chest X-ray interpretation. The post Introducing CARE-X: Towards Clinically Useful Radiology VLMs with Auxiliary Supervision, Reward-Aligned Learning, and Tool-Augmented Measurement appeared first on Microsoft Research.

## Orchard: An open framework for scalable agentic AI

DevFeed: [Orchard: An open framework for scalable agentic AI](<https://devfeed.tech/articles/orchard-an-open-framework-for-scalable-agentic-ai-6806.md>)

Original publisher: [Read original article](<https://www.microsoft.com/en-us/research/blog/orchard-an-open-framework-for-scalable-agentic-ai/>)

Author: Baolin Peng, Wenlin Yao, Qianhui Wu, Hao Cheng, Jianfeng Gao

Published: 2026-08-03T16:00:00Z

Content type: article

Language: en

Sources: [Microsoft Research](<https://devfeed.tech/sources/microsoft-research.md>)

Topics: [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [AI Research](<https://devfeed.tech/topics/ai-research.md>), [Open Source](<https://devfeed.tech/topics/open-source.md>), [Framework](<https://devfeed.tech/topics/framework.md>), [Kubernetes](<https://devfeed.tech/topics/kubernetes.md>), [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>), [computer-use](<https://devfeed.tech/topics/computer-use.md>), [codex](<https://devfeed.tech/topics/codex.md>), [OpenClaw](<https://devfeed.tech/topics/openclaw.md>)

Tags: [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-ai](<https://devfeed.tech/tags/agentic-ai.md>), [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [autonomous](<https://devfeed.tech/tags/autonomous.md>), [autonomous-agents](<https://devfeed.tech/tags/autonomous-agents.md>), [building](<https://devfeed.tech/tags/building.md>), [codex](<https://devfeed.tech/tags/codex.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [framework](<https://devfeed.tech/tags/framework.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [kubernetes](<https://devfeed.tech/tags/kubernetes.md>), [learning](<https://devfeed.tech/tags/learning.md>), [models](<https://devfeed.tech/tags/models.md>), [open](<https://devfeed.tech/tags/open.md>), [open-source](<https://devfeed.tech/tags/open-source.md>), [research](<https://devfeed.tech/tags/research.md>), [research-blog](<https://devfeed.tech/tags/research-blog.md>), [train](<https://devfeed.tech/tags/train.md>), [workflows](<https://devfeed.tech/tags/workflows.md>)

### AI overview

Orchard is an open-source framework for training and evaluating agentic AI systems across software engineering, web navigation, and personal-assistant tasks. Its reusable Orchard Env provides Kubernetes-based infrastructure for data collection, reinforcement-learning rollouts, and evaluation, while Orchard-SWE, Orchard-GUI, and Orchard-Claw demonstrate strong results from relatively small open-weight models.

### Source excerpt

Orchard is an open-source framework for the research community to train and evaluate AI agents across task types. It reduces complexity while supporting strong performance from smaller models by enabling researchers to reuse the same infrastructure. The post Orchard: An open framework for scalable agentic AI appeared first on Microsoft Research.

## Echoverse: Deep, evolving environments for computer-use agents

DevFeed: [Echoverse: Deep, evolving environments for computer-use agents](<https://devfeed.tech/articles/echoverse-deep-evolving-environments-for-computer-use-agents-6787.md>)

Original publisher: [Read original article](<https://www.microsoft.com/en-us/research/blog/echoverse-deep-evolving-environments-for-computer-use-agents/>)

Author: Akshay Nambi, Yash Pandya, Sahil Gupta, Sarthak Harne, Kavyansh Chourasia, Yash Lara, Ahmed Awadallah

Published: 2026-07-30T17:00:00Z

Content type: article

Language: en

Sources: [Microsoft Research](<https://devfeed.tech/sources/microsoft-research.md>)

Topics: [computer-use](<https://devfeed.tech/topics/computer-use.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>), [dataset](<https://devfeed.tech/topics/dataset.md>), [Microsoft](<https://devfeed.tech/topics/microsoft.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [computer-use](<https://devfeed.tech/tags/computer-use.md>), [data](<https://devfeed.tech/tags/data.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [microsoft](<https://devfeed.tech/tags/microsoft.md>), [reinforcement-learning](<https://devfeed.tech/tags/reinforcement-learning.md>), [research](<https://devfeed.tech/tags/research.md>), [research-blog](<https://devfeed.tech/tags/research-blog.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

Echoverse presents twelve high-fidelity training worlds for computer-use agents, designed around realistic application behavior, coherent state, seeded data, and challenging interface controls. Training a 9B model on these environments substantially improved its score, while reinforcement learning with grounded verification helped it generalize and complete goals in fewer steps. Four worlds are released with code, data, and graders to support research.

### Source excerpt

Computer-use AI agents struggle with multi-step workflows like email and customer support. Echoverse trains agents in realistic environments rather than simply providing more training tasks, helping them improve as the tasks, tests, and environments evolve. The post Echoverse: Deep, evolving environments for computer-use agents appeared first on Microsoft Research.

[Next page](<https://devfeed.tech/sources/microsoft-research.md?cursor=WyIyMDI2LTA3LTMwVDE3OjAwOjAwKzAwOjAwIiwgIjExMjcxYTA5LWUwYzMtNDA2Zi1iODgwLTQ0NGZiZDJhZWQzMCJd>)