Zero-shot learning: complete definition in AI for SMEs

Zero-shot learning is a language model's ability to perform a task without any specific examples in the prompt. You simply describe the task and expected output format, and the model executes based on its general knowledge acquired during training. It is the simplest form of prompt engineering: no examples, just instructions.

What it changes for an SME

Zero-shot is the natural starting point for testing an idea:

• "Summarize this contract in 5 key points" → the LLM does it without prior examples;
• "Classify this ticket as urgent, normal or low" → the model understands the categories;
• "Generate 3 variations of this follow-up email" → the format is explicit.

When it is enough and when it is not

Zero-shot works well for simple tasks and powerful models (GPT-4o, Claude). For complex tasks or lighter models, you need few-shot (a few examples). Zero-shot is the quick feasibility test: if it does not work zero-shot, few-shot is the next step. If it still does not work, you need to rethink the task.