RL Scaling Research Scientist — Frontier Models

Speedrun Talent Network; Menlo Park (CA); On-site; USD 225,000 - 350,000; Full time

Job summary

Periodic Labs in Menlo Park (CA) or Montreal seeks a researcher to advance frontier RL for scientific tasks. You will design experiments to understand how RL scales with compute, model size, data, and reward quality, and develop methods from controlled experiments to large-scale runs like Periodic Neon.

This role requires hands-on experience training LLMs with reinforcement learning, a meticulous, scientific approach, and the ability to prototype small-scale RL setups that transfer to bigger

Qualifications

• Hands-on experience training LLMs with reinforcement learning.
• Detail-oriented, rigorous scientific approach.
• Ability to design small-scale RL experiments transferable to large-scale runs.
• Experience debugging and testing research ideas on a complex training stack.

Responsibilities

• Design experiments to understand RL scaling with compute, model size, data, and reward quality.
• Develop RL algorithms across policy optimization, advantage estimation, exploration, and credit assignment.
• Build adaptive sampling and curriculum methods adjusting task difficulty and rollout counts.
• Study bias and stability during RL training and address policy staleness and train–inference mismatch.
• Improve compute efficiency with hyperparameters, length penalties, and update schedules.

Skills

Reinforcement learning

LLM training

Attention to detail

Small-scale RL experiments

Research experimentation

Education

Bachelor's degree

Job description

Periodic Labs in Menlo Park (CA) or Montreal seeks a researcher to advance frontier RL for scientific tasks. You will design experiments to understand how RL scales with compute, model size, data, and reward quality, and develop methods from controlled experiments to large-scale runs like Periodic Neon.

This role requires hands-on experience training LLMs with reinforcement learning, a meticulous, scientific approach, and the ability to prototype small-scale RL setups that transfer to bigger