Zero-shot prompting gives a model an instruction without task-specific example demonstrations; it may still include context. It tests whether the model can perform the task using what it learned during pretraining and post-training alignment. section 2.7.1
Sentiment and customer support
An instruction to classify “The film had poor acting and a dull plot” can produce “Negative.” By contrast, a customer asking about a wrong item may receive generic return advice if the prompt lacks the company’s return policy context. section 2.7.1
Zero-shot prompting can struggle with complex tasks and produce inconsistent results. Add examples when instructions alone fail, provide relevant context for customer-service queries, and test prompts on the target model; wording, formatting, and parameters may need adjustment across models or versions. section 2.7.1
