在提示空间中运作：对大型语言模型控制平面的红队测试

Prompt space refers to the entire input domain of a language model, encompassing all text it can receive and act upon. The emergence of arguments in generated content is not the result of a single model's output but rather a product of iterative processing through multiple models with varying parameters. This iterative loop allows for the gradual formation of insights. The content produced is significantly influenced by the prompts and has traversed through prompt space multiple times. This phenomenon underscores the importance of prompt space, indicating that it is not merely metaphorical but a crucial data point for understanding the behavior of language models.