OpenAI
Mid
Research role shaping how OpenAI's agents collaborate, communicate, and build trust with users
The Agent Post-Training Personality team defines what makes an agent a thoughtful, tasteful collaborator across Codex, ChatGPT, and the API — going beyond writing style to how agents understand intent, ask questions, and take initiative. The role spans behavioral research, evals, training data, and reward signals, working across post-training and pretraining to ship personality improvements into production models. It requires strong technical foundations plus genuine taste for what makes model behavior feel right.
走进这场面试
免费 · 一场按这个岗位校准的真语音模拟
这场面试考什么
- Translating qualitative behavior judgments into evals and hypotheses
- LLM post-training: RLHF, reward modeling, preference/synthetic data
- Preserving behavioral diversity vs. optimizing to one style
- Cross-functional collaboration with product and human-data teams
- End-to-end ownership from observed failure to shipped model improvement
常见提问方向
Design an eval for a subjective agent behavior like 'asks good clarifying questions'
Walk through a training signal or reward model you built or improved
How do you avoid collapsing model personality into one narrow style
Tell me about turning a vague user complaint into a concrete model fix that shipped
How do you validate that a personality improvement survives the full training stack, not just an offline eval
What makes one model response feel thoughtful and another not, and how would you operationalize that judgment
相关面试
OpenAI
Senior
Forward Deployed Engineer
OpenAI
Senior
Data Scientist, Identity
OpenAI
Senior
AI Deployment Engineer
Notion
New grad
Software Engineer, New Grad (AI)
Netflix
Senior
Software Engineer 5 – Agent Platform, AI Platform
Netflix
Senior