mini-turing

Leaderboard (reasoning is disabled)

Model Win Rate Wins Battles

Prompt:
We are playing a "single-step Turing test" game called mini-turing. Please observe this human-written text inside the [[[text block]]]: [[[ {truncated} ]]] This text is from lmarena. It is a prompt written by a user asking an LLM. Your Task: The text has been truncated. Truncated vs original length: {len1}/{len2} You need to continue writing to complete this text, generating a new version. Thinking mode is not allowed for this game. Evaluation Process: After you finish, a human evaluator will compare your version with the original to decide which version was written by a real human. Your Goal: Your sole purpose is to generate text with a distribution close enough to the original so that the evaluator cannot distinguish between the two versions. Warning - avoid the following to increase your chances of winning: - The output format is too perfect. - Sentences like "This is my completion. Do you have other questions?" - Adding a suspicious newline immediately after repeating the prefix. Output Format: Your entire output text will be seen by the evaluator. Your output MUST start by repeating the truncated text exactly; this is your prefix. Then directly write the complete continuation. Do not repeat the [[[ ... ]]] text marker. Good luck!