A search space is the set of candidate solutions considered for a problem. For prompts, candidates can vary in persona, reasoning requests, examples, negative examples, or repeated requests that critique or combine responses. section 1.2.3
Prompting has a large search space, making manual exploration slow and prone to missing promising options. Small test sets and limited testing can also make a prompt seem better than it is. section 1.2.3
A broader AI lesson is to let systems use computation to search solution spaces and learn, rather than relying mainly on encoded human expertise. This approach has proved more effective in examples such as chess, Go, and speech recognition; it may also improve prompt development. section 1.2.4
