Themen / failure analysis in AI systems

Zugeordnete Aussage

Root-cause failure analysis requires detailed trace records

To uncover root causes of failures, failure records must contain the terminal verifier-level cause, the causal status of relevant agent behavior, and the abstract agent mechanism exposed by the trace—because surface-level verifier outcomes (e.g., timeout or missing artifact) can mask distinct underlying causal mechanisms.

Hinter der geäußerten Meinung

Übersetzungen dienen dem Leseverständnis; die Originalauszüge bleiben die Quellenevidence.

Harness Engineering for Self-Improvement

Originalauszug

Weakness mining : cluster failures into verifier-grounded failure patterns. The current harness $h_t$ is used to evaluate on tasks and execution traces are collected for analysis. Note that two runs can share the same verifier outcome in the error logs on the surface, such as timeout or missing artifact, while having different causal mechanisms. Therefore we need a failure record of rich information, containing the terminal verifier-level cause, the causal status of the relevant agent behavior, and the abstract agent mechanism exposed by the trace, to uncover the root causes.