Zero-Shot

Zero-shot means the model can perform a task without any examples. You simply describe what you want: "Classify this email as spam or not spam" or "Translate this sentence to Spanish." The model uses its training knowledge to complete the task without needing demonstrations.

Zero-shot capability is what makes AI tools immediately useful. You do not need to train or configure anything - you just ask. The better a model is at zero-shot tasks, the more useful it is out of the box. Modern frontier models (GPT-4, Claude, Gemini) have remarkably strong zero-shot performance across thousands of tasks.

Brown et al., "Language Models are Few-Shot Learners" - https://arxiv.org/abs/2005.14165

Kojima et al., "Large Language Models are Zero-Shot Reasoners" - https://arxiv.org/abs/2205.11916