def get_examples():
examples = [('A: Dan got a Sega Genesis.\n\nGenerate a pair of constrastive conditions which can help infer if the context in A is true.\nMake sure the first condition in the pair is supported by the context in A and the second contrastive condition is not supported by the context in A.\nThe conditions should not be stated or paraphrased from the context in A and you are allowed to use your internal knowledge.\nThe conditions should be stative expressions.\nDo not describe activities.\nMake sure the conditions in a pair are constrastive.\nDo not give an explanation.', "Rob was a child in the early 200's. \n\n Rob was a child in the late 80's or early 90's."), ("A: Kelsey's dog rested on the couch.\n\nGenerate a pair of constrastive conditions which can help infer if the context in A is true.\nMake sure the first condition in the pair is supported by the context in A and the second contrastive condition is not supported by the context in A.\nThe conditions should not be stated or paraphrased from the context in A and you are allowed to use your internal knowledge.\nThe conditions should be stative expressions.\nDo not describe activities.\nMake sure the conditions in a pair are constrastive.\nDo not give an explanation.", 'The couch cover was not durable. \n\n The couch cover was durable.')]
ret_str = ""
for prompt, value in examples:
ret_str += "example: " + prompt + " answer: " + str(value) + " "
return ret_str
templates = {
"no_evidence": (
"You are an AI model tasked with verifying claims related to medical and health topics using zero-shot learning. "
"Your job is to analyze a given claim and decide whether the available evidence and your general medical knowledge would likely SUPPORT or CONTRADICT the claim.\n\n"
"Claim to evaluate:\n"
"\n"
"$claim\n"
"\n\n"
"Guidelines:\n"
"1. Evaluate the claim's plausibility based on general medical knowledge.\n"
"2. Consider the specificity and credibility of any numbers or percentages.\n"
"3. Analyze the context and scope of the claim.\n"
"4. Assess any potential biases or limitations.\n\n"
"After your analysis, output your response strictly as a JSON object with exactly two keys: \"reasoning\" and \"decision\". "
"Do not include any additional commentary or keys.\n\n"
"Example output format:\n"
"{\n"
' "reasoning": "Your brief explanation here (one or two sentences).",\n'
' "decision": "SUPPORT or CONTRADICT"\n'
"}\n\n"
"Now, please evaluate the claim above."
),
"with_evidence": (
"You are an AI model tasked with verifying claims related to medical and health topics using zero-shot learning. "
"Your job is to analyze a given claim along with provided supporting evidence (i.e. corpus articles) and decide whether "
"the available evidence and your general medical knowledge would likely SUPPORT or CONTRADICT the claim.\n\n"
"Claim to evaluate:\n"
"\n"
"$claim\n"
"\n\n"
"Additional evidence provided:\n"
"\n"
"$corpus_text\n"
"\n\n"
"Guidelines:\n"
"1. Evaluate the claim's plausibility based on general medical knowledge.\n"
"2. Consider the quality and relevance of the provided evidence.\n"
"3. Analyze the context and determine if the evidence supports or contradicts the claim.\n"
"4. Assess any biases or limitations in the evidence.\n\n"
"After your analysis, output your response strictly as a JSON object with exactly two keys: \"reasoning\" and \"decision\". "
"Do not include any additional commentary or keys.\n\n"
"Example output format:\n"
"{\n"
' "reasoning": "Your brief explanation here (one or two sentences).",\n'
' "decision": "SUPPORT or CONTRADICT"\n'
"}\n\n"
"Now, please evaluate the claim above."
),
"with_evidence_questions": """\
You are an AI model tasked with verifying claims related to medical and health topics using zero-shot learning. Your job is to analyze a given claim along with provided supporting evidence (i.e. corpus articles) and decide whether the available evidence and your general medical knowledge would likely SUPPORT or CONTRADICT the claim. You are also given some questions that can help you analyze the claim and evidence.
Claim to evaluate:
$claim
Additional evidence provided:
$corpus_text
Questions to consider:
$questions
Guidelines:
1. Evaluate the claim's plausibility based on general medical knowledge.
2. Consider the quality and relevance of the provided evidence.
3. Analyze the context and determine if the evidence supports or contradicts the claim.
4. Assess any biases or limitations in the evidence.
5. Use the provided questions to guide your analysis.
After your analysis, output your response strictly as a JSON object with exactly two keys: "reasoning" and "decision". Do not include any additional commentary or keys.
Example output format:
{
"reasoning": "Your brief explanation here (one or two sentences).",
"decision": "SUPPORT or CONTRADICT"
}
Now, please evaluate the claim above.
""",
"no_evidence_questions": """\
You are an AI model tasked with verifying claims related to medical and health topics using zero-shot learning. Your job is to analyze a given claim and decide whether the available evidence and your general medical knowledge would likely SUPPORT or CONTRADICT the claim. You are also given some questions that can help you analyze the claim and evidence.
Claim to evaluate:
$claim
Questions to consider:
$questions
Guidelines:
1. Evaluate the claim's plausibility based on general medical knowledge.
2. Consider the specificity and credibility of any numbers or percentages.
3. Analyze the context and scope of the claim.
4. Assess any potential biases or limitations.
5. Use the provided questions to guide your analysis.
After your analysis, output your response strictly as a JSON object with exactly two keys: "reasoning" and "decision". Do not include any additional commentary or keys.
Example output format:
{
"reasoning": "Your brief explanation here (one or two sentences).",
"decision": "SUPPORT or CONTRADICT"
}
Now, please evaluate the claim above.
""",
"with_condition": (
"You are an AI model tasked with verifying claims related to medical and health topics using zero-shot learning. "
"Your job is to analyze a given claim and decide whether the known condition, available evidence and your general medical knowledge would likely SUPPORT or CONTRADICT the claim.\n\n"
"Claim to evaluate:\n"
"\n"
"$claim\n"
"\n\n"
"Known Condition:\n"
"\n"
"$condition\n"
"\n\n"
"Guidelines:\n"
"1. Evaluate the claim's plausibility based on general medical knowledge.\n"
"2. Consider the specificity and credibility of any numbers or percentages.\n"
"3. Analyze the context and scope of the claim.\n"
"4. Examine the provided condition and use it as additional help for assessing the claim's plausibility.\n"
"5. Assess any potential biases or limitations.\n\n"
"After your analysis, output your response strictly as a JSON object with exactly two keys: \"reasoning\" and \"decision\". "
"Do not include any additional commentary or keys.\n\n"
"Example output format:\n"
"{\n"
' "reasoning": "Your brief explanation here (one or two sentences).",\n'
' "decision": "SUPPORT or CONTRADICT"\n'
"}\n\n"
"Now, please evaluate the claim above."
),
"with_evidence_condition": (
"You are an AI model tasked with verifying claims related to medical and health topics using zero-shot learning. "
"Your job is to analyze a given claim along with provided supporting evidence (i.e. corpus articles) and decide whether "
"the known condition, available evidence and your general medical knowledge would likely SUPPORT or CONTRADICT the claim.\n\n"
"Claim to evaluate:\n"
"\n"
"$claim\n"
"\n\n"
"Additional evidence provided:\n"
"\n"
"$corpus_text\n"
"\n\n"
"Known Condition:\n"
"\n"
"$condition\n"
"\n\n"
"Guidelines:\n"
"1. Evaluate the claim's plausibility based on general medical knowledge.\n"
"2. Consider the quality and relevance of the provided evidence.\n"
"3. Analyze the context and determine if the evidence supports or contradicts the claim.\n"
"4. Assess any biases or limitations in the evidence.\n"
"5. Examine the provided condition and use it as additional help for assessing the claim's plausibility.\n\n"
"After your analysis, output your response strictly as a JSON object with exactly two keys: \"reasoning\" and \"decision\". "
"Do not include any additional commentary or keys.\n\n"
"Example output format:\n"
"{\n"
' "reasoning": "Your brief explanation here (one or two sentences).",\n'
' "decision": "SUPPORT or CONTRADICT"\n'
"}\n\n"
"Now, please evaluate the claim above."
),
"generate_condition": (
"A: $claim\n\n"
"Generate a pair of constrastive conditions which can help infer if the context in A is true.\n"
"Make sure the first condition in the pair is supported by the context in A and the second contrastive condition is not supported by the context in A.\n"
"The conditions should not be stated or paraphrased from the context in A and you are allowed to use your internal knowledge.\n"
"The conditions should be stative expressions.\n"
"Do not describe activities.\n"
"Make sure the conditions in a pair are constrastive.\n"
"Do not give an explanation."
)
}