Zero-Shot Learning
/ˈzɪroʊ ʃɒt ˈlɜːrnɪŋ/
A model's ability to perform tasks it wasn't explicitly trained on, using only task descriptions or examples.
Zero-shot learning leverages pre-trained knowledge to handle new tasks without additional training. Large language models excel at this through prompt engineering.
Examples
- GPT-3 translating to languages not in training
- CLIP matching images to unseen concepts
- ChatGPT solving novel reasoning tasks