Machine Learning Scientist

Designs and conducts experiments to evaluate AI models using human preference data, develops new metrics and methodologies, and analyzes large-scale interaction data to advance model reliability and alignment.

About the job

Responsibilities

Design and conduct experiments to evaluate AI model behavior across reasoning, style, robustness, and user preference dimensions
Develop new metrics, methodologies, and evaluation protocols that go beyond traditional benchmarks
Analyze large-scale human voting and interaction data to uncover insights into model performance and user preferences
Collaborate with engineers to implement and scale research findings into production systems
Prototype and test research ideas rapidly, balancing rigor with iteration speed
Author internal reports and external publications that contribute to the broader ML research community
Partner with model providers to shape evaluation questions and support responsible model testing
Contribute to the scientific integrity and transparency of the Arena Intelligence leaderboard and tools

Requirements

Hands-on experience training large-scale models, including reward models, preference models, and fine-tuning LLMs with methods like RLHF, DPO, and contrastive learning
Strong foundation in ML and statistics, with a track record of designing novel training objectives, evaluation schemes, or statistical frameworks to improve model reliability and alignment
Fluent in the full experimental stack, from dataset design and large-batch training to rigorous evaluation and ablation, with an eye for what scales to production
Deeply collaborative mindset, working closely with engineers to productionize research insights and iterating with product teams to align modeling goals with user needs
PhD or equivalent research experience in Machine Learning, Natural Language Processing, Statistics, or a related field
Strong understanding of LLMs and modern deep learning architectures (e.g., Transformers, diffusion models, reinforcement learning with human feedback)
Proficiency in Python and ML research libraries such as PyTorch, JAX, or TensorFlow
Demonstrated ability to design and analyze experiments with statistical rigor
Experience publishing research or working on open-source projects in ML, NLP, or AI evaluation
Comfortable working with real-world usage data and designing metrics beyond standard benchmarks
Ability to translate research questions into practical systems and collaborate across engineering and product teams