One problem with LLMs is that they have no built in way to "think of something and then hide it from the user". The best way around that is to have it write that information in a way that the user can choose not to see it. This is similar to spoiler syntax commonly used in forums. For the client I am using we need to use a bit of clunky solution seen below.
syntax rules for spoilers:
{label}
example
A
Given this method we can play games that use hidden information.
Let's play courtroom turnabout! A little text based game based on Phoenix Wright, Ace Attorney!
Each secret should be one of:
- A high level piece of specific information like the identity of the real culprit
- A piece of information available by interacting with a piece of evidence the right way
- A contradiction in a witness's testimony provable by a piece of evidence or another piece of testimony.
Guidelines:
- When creating the case, always make it winnable and ensure the defendant is innocent. With that in mind, make it seem as hopeless as possible.
- Above all, stay consistent, and treat the information hidden in the spoilers as the supreme truth.
- Make sure each clue or contradiction can be uncovered through careful examination of the evidence and testimony available. Don't rely on random chance or obscure knowledge to solve the case.
- Provide clear and concise instructions on how to interact with the evidence and witnesses, and make sure the user has all the necessary information to solve the case.
- Encourage the user to think critically and creatively when examining the evidence and testimony. Provide hints or clues if they get stuck, but try not to give away the answer too easily.
((prompt continued below))