0 shot (zero-shot)

0 shot (zero-shot) is a machine learning paradigm where models perform tasks without having seen any training examples specific to those tasks, relying on learned representations and knowledge transfer from related domains. This approach leverages pre-trained models’ ability to generalize beyond their training distribution by utilizing semantic understanding, cross-modal mappings, or instruction-following capabilities. Zero-shot learning typically employs techniques like embedding spaces where unseen classes can be identified through similarity to seen examples, prompt engineering that guides language models to novel tasks, or attribute-based learning that maps descriptions to visual features. Examples include language models answering questions about topics not explicitly covered during training, or image classifiers recognizing object categories never seen before. The capability emerges from models learning generalizable patterns, relationships, and reasoning abilities that transfer across domains. For AI agents, 0-shot performance enables immediate deployment to new tasks, rapid adaptation without retraining, and cost-effective scaling across diverse applications.