10 papers
Hardening Agent Benchmarks with Adversarial Hacker-Fixer Loops
Ziqian Zhong, Ivgeni Segal, Ivan Bercovich +3
Agent benchmarks score submissions with outcome verifiers that are typically hand-written and brittle, leaving them open to reward hacking. We audit 1,968 tasks across five termina…
Self-Trained Verification for Training- and Test-Time Self-Improvement
Chen Henry Wu, Aditi Raghunathan
Self-improvement at scale has been a longstanding goal for reasoning models, and there are two natural places to do it: at test time, through verification-refinement (V-R) loops; a…
Understanding and Mitigating Premature Confidence for Better LLM Reasoning
Jingchu Gai, Guanning Zeng, Christina Baek +4
Long chains of thought (CoT) from current language models frequently contain logical gaps and unjustified leaps, limiting the gains from additional test-time compute. Improving rea…
Watch the Weights: Unsupervised monitoring and control of fine-tuned LLMs
Ziqian Zhong, Aditi Raghunathan
The releases of powerful open-weight large language models (LLMs) are often not accompanied by access to their full training data. Existing interpretability methods, particularly t…
Terminal Wrench: A Dataset of 331 Reward-Hackable Environments and 3,632 Exploit Trajectories
Ivan Bercovich, Ivgeni Segal, Kexun Zhang +3
We release Terminal Wrench, a subset of 331 terminal-agent benchmark environments, copied from the popular open benchmarks that are demonstrably reward-hackable. The data set inclu…
Hodoscope: Unsupervised Monitoring for AI Misbehaviors
Ziqian Zhong, Shashwat Saxena, Aditi Raghunathan
Existing approaches to monitoring AI agents rely on supervised evaluation: human-written rules or LLM-based judges that check for known failure modes. However, novel misbehaviors m…