Member of Technical Staff

The Microsoft AI Superintelligence Post Training team is dedicated to advancing post-training methods for both OpenAI and open-source models. Their work encompasses continual pre-training, large-scale deep reinforcement learning running on extensive GPU resources, and significant efforts to curate and synthesize training data. In addition, the team employs various fine-tuning approaches to support both research and product development. The team also develops advanced AI technologies that integrate language and multi-modality for a range of Microsoft products. The team is particularly active in developing code-specific models, including those used in Github Copilot and Visual Studio Code, such as code completion model and the software engineering (SWE) agent models. The team has also produced publications as by-products, including work such as LoRA, DeBerTa, Oscar, Rho-1, Florence, and the open-source Phi models.