Research Fellowship - Open Endedness

About V_max

V_max is an applied research lab working at the frontier of reinforcement learning (RL). We are building new techniques for leveraging RL with Large Language Models (LLMs). Our research contributes directly to our RL platform, which automates the engineering involved in converting data and evals into RL environments.

About the role

This position is for a 6-month research sprint with the V_max team to make progress on our environment generation techniques.

Our goal is to automate the design of tasks for RL agents to help them learn domain specific skills. In this role you will develop approaches to optimize the construction of tasks to maximize resulting agent performance.

Responsibilities

• Develop new approaches to task construction - building on literature in open endedness and unsupervised environment design
• Develop new reward functions for environment design
• Benchmark the agents that learn in generated environments
• Validate your research on industry specific problems

Role Requirements

• Currently enrolled in a PhD, or equivalent experience
• Track record of research excellence, as demonstrated by publications, open source work or publicly deployed AI systems
• Deep understanding of RL and ML
• Significant engineering experience - our research feeds directly into environments and agents that need to be deployed for customers
• Expertise with Python and a ML framework (PyTorch, JAX) is required for this role as well as experience with post-training frameworks

Nice to have

• Experience in post-training LLMs
• Experience researching evolutionary algorithms
• Experience researching unsupervised environment design
• Skilled in presenting the results and implications of your work to multiple levels of audience

Role specific location policy

• This role is based in our San Francisco office; for exceptional candidates we are willing to consider a hybrid arrangement