Microsoft Research Blog

Research Blog

  1. three conversation bubbles on a blue, purple, and pink gradient background

    Steering at the Frontier: Extending the Power of Prompting 

    December 12, 2023 | Eric Horvitz, Harsha Nori, and Yin Tat Lee

    We’re seeing exciting capabilities of frontier foundation models, including intriguing powers of abstraction, generalization, and composition across numerous areas of knowledge and expertise. Even seasoned AI researchers have been impressed with the ability to steer the models with straightforward, zero-shot prompts. Beyond basic, out-of-the-box prompting, we’ve been…

  2. Satya Nadella on stage at Microsoft Ignite 2023 announcing Phi-2.

    Phi-2: The surprising power of small language models 

    December 12, 2023 | Mojan Javaheripi and Sébastien Bubeck

    Phi-2 is now accessible on the Azure model catalog. Its compact size and new innovations in model scaling and training data curation make it ideal for exploration around mechanistic interpretability, safety improvements, and fine-tuning experimentation on a variety of tasks.

  7. Research Focus Edition 30 December 6, 2023

    Research Focus: Week of December 4, 2023 

    December 6, 2023

    Research Focus: Using LLMs in a Rust-based formal verification framework; Rethinking network measurements with user feedback; 3D telemedicine using HoloportationTM communication technology could enhance overseas surgical visits.

  9. Blue to green gradient. Two rows of hands: the top row signing ASL and the bottom row signing Data.

    Tackling sign language data inequity 

    December 4, 2023

    ASL Citizen is the first crowdsourced sign language dataset, advancing the state of the art in sign recognition. The web-based project captured input from people in real-world settings, and from a diverse group of experts, including Deaf team members.

  11. Illustrated icons of a medical bag, hexagon with circles at its points, and a chat bubble on a blue and purple gradient background.

    The Power of Prompting 

    November 28, 2023 | Eric Horvitz

    Microsoft Chief Scientific Officer Eric Horvitz explains how new prompting strategies can enable generalist large language models like GPT-4 to achieve exceptional expertise in specific domains, such as medicine, and outperform fine-tuned specialist models.

  12. EMNLP 2023 blog hero - female radiologist analyzing an MRI image of the head

    GPT-4’s potential in shaping the future of radiology 

    November 27, 2023 | Javier Alvarez-Valle and Matthew Lungren

    This research paper is being presented at the 2023 Conference on Empirical Methods in Natural Language Processing (opens in new tab) (EMNLP 2023), the premier conference on natural language processing and artificial intelligence. In recent years, AI has been increasingly integrated into healthcare, bringing about…