CyberAgent训练与调试专家 - 安全与风控
Core
Build and debug safety systems for large language models (LLMs) by identifying capability gaps, designing training tasks, and creating automated validation pipelines to ensure security across training, inference, and deployment.
Role type
Senior LLM Safety Engineer (Training & Debugging)
Builds
Automated validation pipelines, high-quality training data, and safety evaluation frameworks for LLMs.
Domain
Artificial Intelligence / Cybersecurity
Deliverable
production ML models
Required skills
System programming and debugging (C/C++), Python, Linux, containerization, virtualization, experimental design, root cause analysis, tool-use trajectory analysis, engineering implementation.
Preferred skills
LLM Agent development, SFT, RL, Reward Design, Tool-use, data annotation systems, automated evaluation.
Responsibilities
Identify model capability gaps and translate them into training tasks and evaluation signals; Analyze root causes of model failures in reasoning, experimentation, or tool usage; Design reproducible experiments with clear success/failure criteria; Collaborate with training teams to supplement high-quality data; Establish automated verification to distinguish valid from invalid model behaviors; Conduct joint experiments and independent evaluations with model training teams.
