Researcher (Engineer/Scientist), Training Architecture

OpenAI is seeking a Researcher (Engineer/Scientist) for their Training Architecture team in San Francisco. This role is crucial in pushing the boundaries of large language model (LLM) architectures, enhancing intelligence, efficiency, and adding new capabilities to OpenAI's flagship models.

The ideal candidate will have a deep understanding of LLM architectures, sophisticated knowledge of model inference, and a hands-on empirical approach. They should be comfortable with creative breakthroughs, strengthening baselines, designing evaluations, debugging regressions, and identifying bottlenecks.

As part of the Training team, you'll be responsible for producing large language models that power OpenAI's research, products, and ultimately bring the company closer to AGI. The role involves combining deep research to improve current architecture, datasets, and optimization techniques, alongside long-term strategies to enhance the efficiency and capability of future model generations.

Key responsibilities include designing and scaling new architectures, executing and analyzing experiments, optimizing model and computational performance, and contributing to training and inference infrastructure. The position requires experience with major LLM training runs, self-directed evaluation and improvement of deep learning architectures, and a strong motivation for safely deploying LLMs in real-world scenarios.

OpenAI offers a competitive salary range of $360K – $440K, along with generous equity and benefits. The company provides comprehensive health insurance, mental health support, a 401(k) plan with matching, unlimited time off, paid parental leave, and an annual learning stipend.

This is a hybrid role based in San Francisco, requiring 3 days in the office per week. OpenAI is committed to diversity, equal opportunity, and creating AI systems with safety and human needs at their core. Join OpenAI in shaping the future of AI technology and ensuring its benefits are widely shared across humanity.