Research Engineer, Value Persistence Through Reinforcement Learning

Remote, Global

$9,167 per month

• In this role, you'll research whether values instilled through midtraining persist after reinforcement learning.
• Build and run training pipelines combining midtraining on synthetic corpora with RLVR or GRPO post-training.
• Implement synthetic corpus generation with well-matched controls at required scale and quality.
• Run evaluations and interpretability methods to understand value persistence and erosion mechanisms.
• Maintain reproducible configurations, logging, and experiment tracking while co-authoring papers.

Compassion Aligned Machine Learning conducts research on aligning AI with the well-being of all sentient beings.