PeopleofSuperintelligence

Researcher

Corby Rosset

Also known as Corbin Rosset

Corby Rosset is a Microsoft researcher whose work spans web search, language models, and computer-use agents. He helped develop search-intent systems and wrote about Turing-NLG before coauthoring Orca 2 and Orca-Math, research on reasoning and math in smaller models. He presented Direct Nash Optimization and coauthored a dataset of complex search questions. As of October 2026, he is a researcher at Microsoft whose recent work includes the Universal Verifier and Fara1.5.

Now at
Microsoft
Works
30
Citations
665
h-index
12
Badges

Publication numbers from OpenAlex, as of 19 October 2026.

Timeline

  1. 2017

    Computer science at Johns Hopkins

    In 2017, he is a senior in Johns Hopkins computer science and describes machine learning as a field where interdisciplinary students can have an impact.

  2. 2018–2019

    Search intent and retrieval

    In 2018, he coauthors a reinforcement-learning approach to Bing search candidate generation that reduces index blocks accessed by up to 20%. By 2019, he is an Applied Scientist at Microsoft Bing and helps develop MS GEN Encoder, which maps queries with similar click results to similar representations.

  3. 2020

    Turing-NLG at 17 billion

    In 2020, he writes about T-NLG, Microsoft's 17-billion-parameter language model for text generation, question answering, and summarization.

  4. 2023

    Orca 2 reasoning models

    In 2023, he coauthors Orca 2, a 7B and 13B model family fine-tuned to use varied strategies for reasoning tasks.

  5. 2024

    Math models and preference learning

    In 2024, he coauthors Orca-Math, a 7-billion-parameter model for grade-school math, and presents Direct Nash Optimization, which teaches language models to improve through general preferences. He also coauthors Researchy Questions, a dataset of search queries filtered for non-factoid, decompositional, and multi-perspective questions.

  6. 2025

    A model for computer use

    In 2025, he is a Senior Researcher at Microsoft and coauthors Fara-7B, the company's first agentic small language model designed for computer use.

  7. As of October 2026

    Researcher at Microsoft

    In 2026, he coauthors a paper on the Universal Verifier for checking web-agent task outcomes and contributes to Fara1.5, a small computer-use model family for browser tasks. As of October 2026, he works as a researcher at Microsoft on computer-use agents.

Career

  1. Since 2025MicrosoftApplied Scientist

Education

  1. Johns Hopkins University

From the biography and OpenAlex affiliations (the years of papers that list the organization), so some years are approximate.

Most-cited works

  1. Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone2024 · 164 citationsWith Adil Salim, Ahmed Awadallah, Caio César Teodoro Mendes, Ce Liu, Ronen Eldan and Shital Shah
  2. Leading Conversational Search by Suggesting Useful Questions2020 · 100 citations
  3. Transformer-XH: Multi-Evidence Reasoning with eXtra Hop Attention2020 · International Conference on Learning Representations · 99 citations
  4. LLM-Rubric: A Multidimensional, Calibrated Approach to Automated Evaluation of Natural Language Texts2024 · Annual Meeting of the Association for Computational Linguistics (ACL) · 46 citationsWith Jason Eisner
  5. Knowledge-Aware Language Model Pretraining2020 · 44 citations

From OpenAlex, as of 19 October 2026.

  1. The company share of AI researchers’ top papers rose from 2% to 32%Includes Microsoft. We estimate where leading AI researchers worked when their most-cited papers appeared. The company share rose from 2% in 1975–1979 to 32% in 2020–2024.
  2. Google has the highest combined h-index among AI organizationsIncludes Microsoft. Our estimates of combined and median h-index at AI organizations, using current affiliations and OpenAlex records.
  3. Google alumni founded or led 87 other organizations, the most among the labs we compareIncludes Microsoft. We trace 212 organizations founded or led by AI lab alumni, and 223 career moves between the labs.
  4. Google had the most people shaping AI every year from 2014 to 2025Includes Microsoft. Our estimates of which organizations employed the most people shaping AI from 1961 to 2025, with a year-by-year chart.

Colleagues

People with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Corby Rosset yet. Send us a free-licensed photo.

  1. Why CS? | Department of Computer Science Turns 30 cs.jhu.edu. Checked 19 October 2026.
  2. Optimizing Query Evaluations using Reinforcement Learning for Web Search arxiv.org. Checked 19 October 2026.
  3. Learning web search intent representations from massive web search logs microsoft.com. Checked 19 October 2026.
  4. Turing-NLG: A 17-billion-parameter language model by Microsoft microsoft.com. Checked 19 October 2026.
  5. Orca 2: Teaching Small Language Models How to Reason microsoft.com. Checked 19 October 2026.
  6. Orca-Math: Demonstrating the potential of SLMs with model specialization microsoft.com. Checked 19 October 2026.
  7. Direct Nash Optimization: Teaching language models to self-improve with general preferences microsoft.com. Checked 19 October 2026.
  8. Researchy Questions: A Dataset of Multi-Perspective, Decompositional Questions for LLM Web Agents arxiv.org. Checked 19 October 2026.
  9. Fara-7B: An Efficient Agentic Model for Computer Use microsoft.com. Checked 19 October 2026.
  10. The Art of Building Verifiers for Computer Use Agents microsoft.com. Checked 19 October 2026.
  11. MagenticLite, MagenticBrain, Fara1.5: An agentic experience optimized for small models microsoft.com. Checked 19 October 2026.