About the Role
When you look at the responses from today's leading language models, do you wonder, "How do we align these systems with human values and preferences?" or “How can we improve an LLM’s abilities beyond what a human can achieve?”
The Reward Modeling team at Anthropic is working to develop sophisticated techniques for teaching AI systems to understand and embody human values, as well as to push forward AI capabilities. We believe that robust reward models are critical to training AI systems that advance the frontier of safety and capabilities. We're looking for engineers to join our efforts to push forward the science of reward modeling
Responsibilities:
- Help implement novel reward modeling architectures and techniques
- Optimize training pipelines
- Build and optimize data pipelines
- Collaborate across teams to integrate reward modeling advances into production systems
- Communicate engineering progress through internal documentation and potential publications
You may be a good fit if you:
- Have a strong engineering background in machine learning, with demonstrable expertise in preference learning, reinforcement learning, deep learning, or related areas
- Are proficient in Python, deep learning frameworks, and distributed computing
- Are familiar with modern LLM architectures and alignment techniques
- Have experience with improving model training pipelines and building data pipelines
- Are comfortable with the experimental nature of frontier AI research
- View research and engineering as complementary disciplines and are willing to implement some research ideas
- Can clearly communicate complex technical concepts and research findings
- Have a deep interest in AI alignment and safety
- Proficiency in Python and experience with deep learning frameworks is required for this role
Experience with reward models is not required, but experience with LLMs or other large models is a significant plus. We welcome candidates at various experience levels, with a preference for senior engineers who have hands-on experience with frontier AI systems.
