Zero-shot learning is a model's ability to perform a task it has never been explicitly trained on, with no examples provided in the prompt. The model relies purely on its pre-trained knowledge and generalization. For instance, you can ask an LLM to translate a Darija sentence to Japanese even if no Darija-Japanese translation pair existed in its training data. This flexibility is a hallmark of large language models and is expanding the range of tasks AI can handle without custom training.