Zero-Shot Learning

Zero-shot learning is when a model performs a task it was not explicitly trained or given examples for, relying solely on its pre-trained knowledge and the task instruction. Agent governance must account for the higher unpredictability of zero-shot performance.