Tag
This paper finds that prior audit and repair episodes in context reduce false alarms in LLM verifiers by shifting decision thresholds, with repair content and audit verdict complementarily affecting different model families.
The paper proposes Signal-Driven Observation (SDO), a method for web agents to avoid context degradation by only reading task-relevant parts of the DOM and re-invoking observation only when triggered by specific signals, rather than reading the full page state at every action step.
This paper derives tight theoretical bounds for human-AI teams, proving when confidence-based aggregation leads to complementarity and establishing impossibility results under specific error correlations.