大模型安全与对齐研究工程师 - Seed Model
Core
Researching safety and alignment for large language models, including SFT, preference optimization, RL, agent safety, automated red teaming, and risk monitoring.
Role type
Senior IC research engineer (LLM safety & alignment)
Builds
Safer, aligned large models and agent systems deployed in production
Domain
AI / Machine Learning / LLM Safety
Deliverable
production ML models
Required skills
Machine learning, deep learning, statistics, Python, PyTorch, SFT, reinforcement learning, agent systems, risk modeling, automated red teaming, monitoring mechanisms
Preferred skills
Distributed training, Agent RL, long-horizon tasks, multi-agent training, AI control, mechanism interpretability, CBRN/ cybersecurity evaluation
Technologies
Python, PyTorch
Responsibilities
Research SFT, preference optimization, and RL to reduce deception and overreach; Study agent behavior boundaries in tool use and code execution; Conduct risk assessment and threat modeling for high-risk domains; Build automated attack generation and adversarial training methods; Explore input/output monitoring and permission control mechanisms; Analyze failure cases and apply solutions to large-scale training and deployment
