FIELD NOTES TOPIC
ai debugging
2 notes, newest first.
TRAJDEBUG: Tracing Error Lifecycle to Identify Critical Failures in Long-Horizon Agent Trajectories
The paper proposes TrajDebug, a framework for error-lifecycle tracing in long-trajectory agentic systems. The goal is to identify critical errors responsible for final failures, especially in complex domains where…
TRAJDEBUG: Tracing Error Lifecycle to Identify Critical Failures in Long-Horizon Agent Trajectories
In this study, researchers present TrajDebug, a framework designed to identify critical errors in long trajectories of large language models (LLM) used in complex domains. The framework addresses two challenges: first…
