Researcher, Artifacts - Agent Post-Training

📋 Description

• Design experiments to improve agentic model behavior for software/plugins.
• Own end-to-end post-training stack improvements: RL, data pipelines, graders, evals.
• Build evals/environments to reveal model failures and convert them into training data or fixes.
• Partner with Codex and ChatGPT teams to translate user signals into model improvements.
• Work on early-training and alignment interventions: data mixtures, objectives, eval loops.
• Improve large-scale training and launch machinery: velocity, reliability, observability, cost, latency.

🎯 Requirements

• Strong fundamentals in ML, software, systems, or statistics; quick to learn new areas.
• Hands-on with LLMs, RL, RLHF/RLAIF, post-training, evals, graders, synthetic data, production ML.
• Thrives on open-ended problems with noisy signals; blends research taste and engineering.
• Cares about product impact and practical model behavior.
• Turn vague problems into experiments: hypothesize, build pipeline, run, analyze, decide.
• Comfortable across research, product, infra, data, evals, safety; communicates clearly.