update prompts to make clear the usage of chain-of-thought

tests/config/prompts.json

@@ -1,8 +1,9 @@
 {
     "sample": "Here are some examples that have already been identified as either human-written or AI-written. ",
     "description": "Using the examples presented above, can you describe differences between works written by humans and by AI?",
-    "introduction": "Please help me determine whether the text was made by a human or an LLM:",
-    "classify": "Please help me determine whether this piece text above is written a human or by a large language model.",
-    "judge": "Try to consider both options before coming up with a single-choice conclusion.",
-    "answer format": "Your response must always end with your verdict in a new line that either contains “human” or “AI” without any additional words, quotation marks, formatting, or punctuations. Examples of responses include: \n“ABC \n\nhuman”\nor\n“DEF \n\nAI”\n\nwhere “ABC” and “DEF” are placeholders for your explanations."
-}
+    "bridge": "Now, keep in mind how those examples were classified.",
+    "introduction": "Below is some text that I need help on:",
+    "classify": "Do you think this text was highly likely to be written by a human or by an LLM?",
+    "judge": "Let us evaluate both options before coming up with a conclusion.",
+    "answer format": "Having stated that, please summarize that answer in one word: either “human” or “AI”. "
+}