Zero-shot
Zero-shot refers to the ability of an AI model to understand and perform new tasks directly, without any additional training specific to the task, relying solely on its existing general knowledge.
How it works
This is possible because modern large language models (such as the GPT series) learn from vast and diverse data during training, building strong generalization capabilities. This enables the model to understand new instructions described in natural language by users and generate reasonable responses, greatly improving AI's flexibility and ease of use.
Examples
- 1.Asking an AI model that has never been specifically trained for translation to translate directly based on the instruction "Translate 'Hello, World' into French."
- 2.Showing an image recognition AI a picture of an exotic fruit it has never seen before and asking directly "What fruit is this?" — the model attempts to identify and describe it based on its visual features.
- 3.In customer service scenarios, an AI can directly answer user questions about a new product without being trained on that product's FAQ.