Research Engineer / Research Scientist - Agent Robustness and Control
Description
The Agents team within Research works in research areas combining language models, reinforcement learning, and agents. Agent Robustness and Control focuses on ensuring safe and secure iterative deployment of agents through systems, product, and research engineering work.
About the Role
We’re looking for research engineers excited to advance the frontier of agent research, safety, and security. This role is focused on integrating safety and security at every layer of the stack from fundamental research to launching the work as a product.
This role is based in San Francisco, CA. We use a hybrid work model of 3 days in the office per week and offer relocation assistance to new employees.
In this role, you will:
Train and evaluate models with a focus on robustness and monitoring-in-the-loop
Own and pursue innovative solutions to safety and security risks with no clear answer
Design, implement, test, and debug code across our research and product stack
Collaborate with Safety, Security, Research, and Product teams to ship capable agents
Build reusable tools for improving model safety and robustness across systems and products
Actively evaluate and understand the safety of our models and systems, identifying areas of risk, then proposing and implementing mitigation strategies
Qualifications
Are happy to learn whatever is needed and have unique way of learning quickly
Are comfortable diving into a large codebase to debug (including ML or product components)
Are willing to experiment and be creative in problem solving, and have a strong understanding that there's more than one way to solve any problem
Have strong programming skills in Python or similar languages
Have experience balancing product and research objectives with safety and security considerations in a production, high-stakes environment
Are familiar with methods of training and fine-tuning large language models, such as distillation, supervised fine-tuning, and/or policy optimization