Research Scientist, Multi-Agent

Anthropic•Seattle, WA

9d•$500,000 - $850,000•Hybrid

About The Position

Multi-Agent systems are becoming an increasingly important part of how AI is deployed, whether via fast small-model subagents inside a product, or large groups of agents solving very large problems. Training Claude to be maximally effective and safe within large groups is a challenging new area of reinforcement learning, and represents a new axis for scaling test time compute. We are looking for researchers who have experience training multi-agent systems at the largest scale and an appreciation for the incentives and mechanism design that come into play.

Requirements

Experience with large-scale RL on language models
Experience training multi-agent systems
Enjoy going deeply into the roots of a problem and understanding its foundations, rather than its surface.
Good communication skills and an interest in working with other researchers on difficult tasks
A passion for making powerful technology safe and societally beneficial
Excited for a mission-driven org with fast-paced, impactful work

Responsibilities

Ideate, develop, and compare the performance of different agent harness configurations (eg memory, context management, communication architectures for agents)
Design and implement rigorous quantitative benchmarks for large scale agentic tasks
Work with our product org to find solutions to our most vexing challenges in applying agents to our products
Help create and optimize environments and data for model training that maximize Claude’s performance or ease of use on agentic tasks