The AI won't always get the right response on the first shot. The ability to regenerate a prompt instructs the AI to try again.
In many cases, this generates a new variations, allowing the user to navigate between their options and then choose the best fit. In this way, it can operate as a type of remix input, building off the same tokens but possibly playing with different weights behind the scenes.
An anti-pattern occurs when the user regenerates the prompt but it writes over the first version. In these cases, the user can't take pieces of the prompt that worked and blend it with a new result. The AI has moved their agency (this is the case with Github Copilot's inline prompts).
The first time a user runs a prompt, they see the AI's first best guess of what the most probable output should be that meets the their intent. However, there may be other logic paths the AI might take through the data to get to a similarly good or even better response. Running the prompt multiple times gives users a better sense for how the AI is understanding their prompt, and a better shot to get a good result.
From here, the user can keep regenerating a response, or rework their prompt, add tuners, and constrain the AI to get where you they want it to go.
