Agent Post-Training, Context Research

Highlights

• Scale compute spent on context to enable the next paradigm of model training.
• Train frontier agents shipped in Codex, ChatGPT, API and other products.
• Build evals and environments that turn model failures into training signal.

Ideal candidate

High-agency builder with strong ML fundamentals who thrives on open-ended problems and cares deeply about shipping useful, reliable agents into real products.

Requirements

• Strong technical fundamentals in machine learning, software engineering, systems, statistics or related field
• Hands-on experience with LLMs, RL, post-training, evals, graders or tool-using agents
• Excited by open-ended problems with noisy signal requiring research taste and engineering execution
• Able to move from vague behavioral problems to concrete experiments and fixes
• Comfortable working across research, product, infrastructure, data and safety teams