Research Engineer / Research Scientist, Multimodal

You want to build cutting-edge Artificial Intelligence from the ground up. You are passionate that these systems are safe and trustworthy, and care about their broader societal impact. In this role, you\’ll work on research, development, and infrastructure for state-of-the-art large language models, with a focus on multimodal capabilities. This role will touch all parts of the research ecosystem, from building infrastructure to developing research prototypes to running multi-thousand accelerator training jobs. You may be a good fit if some of the following apply to you: Care deeply about the societal impacts of your work Want to work on the frontier of Artificial Intelligence Have substantial software engineering experience through industry, academia, or other projects Have research experience, through writing scientific publications, or involvement in personal or industrial projects Are results oriented and flexible in your approach Are willing to pick up slack, even if it goes outside your job description Like working in a close-knit team environment, and enjoy pair programming! Strong candidates may have experience with some of the following: High performance, large-scale Machine Learning systems ML hardware, frameworks, and infrastructure, such as TPUs, GPUs, Jax, PyTorch, OS internals, and Kubernetes Language modeling with transformers Deep learning research on images, videos, audio, or other modalities Representative projects: Design a training loss for a new modality Design and run experiments to evaluate the scalability of two architectural variants Analyze and debug a large-scale training run Scale an architecture to optimize throughput on thousands of GPUs Build from-scratch a new deep learning architectural components Build a pipeline to ingest a novel source of data Build a language model evaluation Build a tool for data visualization Review the scientific literature in a domain and write a design doc on how techniques could be applied #J-18808-Ljbffr