[KongchangAI]
Concept错误可验证性

Error Detectability

Refers to the ease with which the output of an AI agent task can be verified, and is a key dimension in evaluating the net benefit of an agent. Errors in tasks with deterministic criteria are exposed immediately, whereas errors in tasks lacking automated criteria remain hidden and costly to repair.

Timeline (last 90 days)

Nov 3

The gap between the oracle upper bound (assuming selected errors can be perfectly corrected) and the actual gain essentially reflects the mismatch between error fixability and error detectability.

Unverified50%

All Facts (1)

Source Articles