Few-shot
Few-shot learning is the ability of AI models to understand and perform new tasks from just a handful of examples.
How it works
It works by including a few task examples in the instructions (prompt) given to the model, letting it learn patterns from them without time-consuming retraining. This approach greatly improves AI's flexibility and usefulness, allowing it to quickly adapt to all kinds of new needs.
Examples
- 1.Teach an AI to summarize the sentiment of new product reviews based on a few example reviews
- 2.Show the AI a few emails with their category labels so it learns to classify new emails automatically
- 3.Provide a few Chinese-English sentence pairs so the LLM grasps the task and starts translating