Difficulty-calibrated task selection
For training, tasks should expose current agent weaknesses—i.e., be feasible and realistic but not yet consistently solved. Reliably solved tasks provide little new training signal.
Éléments favorables
AutoSynthData: Generating Training Data for Enterprise Agents
Extrait original
Difficulty. For training, the task should expose a weakness of the current agent. Tasks that are already solved reliably provide little new training signal. The useful region is therefore tasks that are feasible and realistic, but not yet consistently solved.