About the Role
The role involves developing and applying post-training techniques to enhance the safety, reliability, and performance of large AI models, combining research innovation with production-grade engineering.
Responsibilities
- Develop and refine post-training methods for large-scale language models.
- Implement scalable algorithms to improve model alignment and behavior.
- Collaborate with research teams to transition experimental techniques into production.
- Optimize training pipelines for efficiency and reproducibility.
- Diagnose and resolve issues in model performance during post-training phases.
- Contribute to software infrastructure supporting iterative model refinement.
- Work closely with safety teams to evaluate behavioral changes from post-training updates.
- Design automated evaluation frameworks for model outputs.
- Integrate feedback mechanisms into training loops.
- Ensure compatibility between model updates and deployment environments.
- Monitor system performance across training cycles.
- Document methods and findings for internal knowledge sharing.
- Support version control and experiment tracking systems.
- Contribute to best practices for model fine-tuning and evaluation.
- Maintain high standards for code quality and system reliability.
Nice to Have
- Advanced degree in computer science, machine learning, or related field.
- Prior work on model alignment or safety techniques.
- Experience deploying ML models in production settings.
- Knowledge of reinforcement learning from human feedback (RLHF).
- Familiarity with large-scale training infrastructure.
- Contributions to open-source machine learning projects.
- Research publications in relevant technical areas.
- Experience with automated testing in ML systems.
- Understanding of model evaluation metrics and benchmarks.
- Background in natural language processing tasks.
Compensation
Competitive salary and equity offered based on experience and location.
Work Arrangement
Hybrid work model with flexibility depending on team and role requirements.
Team
Part of a multidisciplinary research and engineering team focused on developing safe and reliable AI systems.
Research Focus
- Focus on advancing post-training methodologies such as fine-tuning, distillation, and reinforcement learning to improve model behavior.
- Explore novel approaches to align models with human intent and safety goals.
- Collaborate on experiments that test the limits of current post-training techniques.
Engineering Impact
- Build tools and systems that enable efficient iteration on model improvements.
- Ensure that research prototypes can scale reliably in production environments.
- Contribute to reusable libraries for model evaluation and training.
Visa sponsorship available for qualified international candidates.
