Inverse Rubric Optimization: A testbed for agent science | Fulcrum
We propose inverse rubric optimization (IRO): tasks where an agent must learn the preferences of a black-box judge under a label budget. IRO tasks induce rich agent behavior and smooth scaling, making them a useful testbed for agent science.
“It is important to draw wisdom from many different places. If you take it from only one place, it becomes rigid and stale.” — Uncle Iroh At Fulcrum Research, we study the performance and behavior of long-horizon agents. Although each task setting has its own specific structure, we believe it’s possible to find general principles of agent performance across settings, each contributing to a nascent agent science. In this post, we motivate the difficulty of finding suitable settings for agent science and propose inverse rubric optimization (IRO) settings, in which an agent has to optimize the pr
saved by
related reading
- Measuring Reward-Seeking by Instilling Contrastive Beliefsalignment.openai.com
- Demystifying evals for AI agents \ Anthropicanthropic.com
- Surfacing Pathological Behaviors in Language Models | Transluce AItransluce.org
- How Claude Performs on Robotics Tasks \ Anthropicanthropic.com
- Scaling Activation Oracles to Trillion-Parameter Modelstransluce.org
- Foundation Models for Oversight | Transluce AItransluce.org
- Quantifying infrastructure noise in agentic coding evals \ Anthropicanthropic.com
- Rubrics as Rewards: Reinforcement Learning Beyond Verifiable Domainsarxiv.org
- Coding vs thinking — Paradigm 3paradigm3.org
- Agent Evaluation: A Detailed Guidecameronrwolfe.substack.com
- Machine Studying | Jacob Xiaochen Lijacobxli.com
- The Verification Horizon: No Silver Bullet for Coding Agent Rewardsarxiv.org