Use gpt4o and apply CoT in prompting

constants.py

@@ -20,7 +20,7 @@ SDG_DETAILS = """
 
 
 GPT_PROMPT_parallel="""
-For AI generated/modified parts, our base model is GPT-3.5-turbo. 
+For AI generated/modified parts, our base model is GPT-4o. 
 
 For direct generation, we use instructions like following:
 "Given the task as: [task_description], provide an answer: "
@@ -33,7 +33,7 @@ Merge the two answers into one in a coherent way: "
 """
 
 GPT_PROMPT_sequential="""
-For AI generated/modified parts, our base model is GPT-3.5-turbo. 
+For AI generated/modified parts, our base model is GPT-4o. 
 
 For the modification, we use instructions like following:
 "Given the task as :[task_description, the human answer is: [human_text]
@@ -41,7 +41,7 @@ Provide an answer of your own but make sure it is coherent and should be based o
 """
 
 GPT_PROMPT_reverse_sequential="""
-For AI generated/modified parts, our base model is GPT-3.5-turbo. 
+For AI generated/modified parts, our base model is GPT-4o. 
 
 For direct generation, we use instructions like following:
 "Given the task as: [task_description], provide an answer: "

utils.py

@@ -31,15 +31,23 @@ def describe_task():
     return task_description
 
 def generate_text_with_gpt(prompts, api_key = None):
-    """Generate text using the GPT-3 model."""
+    """Generate text using the GPT-4o model."""
     if api_key:
         openai.api_key = api_key
 
     try:
         response = openai.ChatCompletion.create(
-            model="gpt-3.5-turbo",
+            model="gpt-4o",
             messages=[
-                {"role": "system", "content": "Please assist."},
+                {"role": "system", "content": """
+                 You are a sustainability expert with an extensive background in SDG17 
+                 (Partnerships for the Goals) and the United Nations Sustainable Development Goals (SDGs). 
+                 Your role is to critically analyze and evaluate the sustainability of business models, 
+                 using a highly rigorous and evidence-based approach. 
+                 You must maintain strict scoring criteria with clear justification for each score, 
+                 ensuring a wide distribution of scores to reflect meaningful differentiation in compliance and impact levels. 
+                 Avoid overly lenient assessments.
+                 """},
                 {"role": "user", "content": prompts}
             ]
         )
@@ -54,41 +62,87 @@ def generate_ai_initial_answer(task_description, api_key=None):
     return generate_text_with_gpt(prompt, api_key)
 
 def merge_texts_parallel(task_description, human_text, ai_text, api_key = None):
-    prompt = f"Given the task: {task_description}, there are two answers provided:\n" + \
-             f"The first answer:  {human_text}\nThe second answer: {ai_text}\n" + \
-             f"Merge the two answers into one in a coherent way: "
+    prompt = (
+    "You are tasked with merging two answers into a single, coherent, and logically structured response. "
+    "Your response should synthesize the strengths of both answers, eliminate redundancy, and ensure a logical flow. "
+    "Follow this structured chain of thought to guide your process:\n\n"
+    "1. **Understand the Task:** Start by carefully analyzing the task description to ensure your final response aligns fully with the objective.\n"
+    "2. **Extract Key Points:** Review both answers and identify the key ideas and arguments they present.\n"
+    "3. **Resolve Redundancies and Conflicts:** Compare the two answers to eliminate repetitive content and resolve any contradictions.\n"
+    "4. **Integrate Seamlessly:** Combine the extracted points into a single, unified response, ensuring a logical and coherent structure.\n"
+    "5. **Refine for Clarity:** Polish the final response for clarity, conciseness, and alignment with the task requirements.\n\n"
+    "Task Description:\n"
+    f"{task_description}\n\n"
+    "Answer 1:\n"
+    f"{human_text}\n\n"
+    "Answer 2:\n"
+    f"{ai_text}\n\n"
+    "Provide the unified and refined response below, marked clearly as the final version."
+)
     return generate_text_with_gpt(prompt, api_key)
 
 def merge_texts_sequential(task_description, human_text, api_key = None):
-    prompt = f"Given the task:{task_description}, here is the answer provided by the human: {human_text}\n" + \
-             f"Refine this response and ensure the final answer aligns with the human's intent:"
+    prompt = (
+    f"Given the task: {task_description}, here is the response provided by the human:\n"
+    f"{human_text}\n\n"
+    "Your task is to refine the human's response while ensuring that the final answer fully aligns with their intent. "
+    "Follow this logical chain of thought:\n"
+    "1. **Understand the Intent:** Analyze the human's response to identify their key points and intended message.\n"
+    "2. **Evaluate Clarity:** Check for any ambiguities, gaps, or unclear phrasing in the response.\n"
+    "3. **Improve Coherence:** Ensure the response flows logically and connects ideas seamlessly.\n"
+    "4. **Enhance Precision:** Refine wording to be concise and impactful while preserving the human's original intent.\n"
+    "5. **Check Alignment:** Confirm that the final response accurately represents and strengthens the human's input without introducing unintended changes.\n\n"
+    "Provide the refined response below, clearly marked as the final version."
+)
+
     return generate_text_with_gpt(prompt, api_key)
 
 def modify_with_suggestion(task_description, text, suggestions, api_key = None):
-    prompt = f"Given the task:{task_description}, the answer provided is: {text}\n" + \
-             f"Modify the answer based on the following suggestions: {suggestions}"
+    prompt = (
+    f"Given the task: {task_description}, the provided answer is:\n"
+    f"{text}\n\n"
+    f"The following suggestions have been made for improvement:\n"
+    f"{suggestions}\n\n"
+    "Your task is to modify the provided answer by systematically incorporating the suggestions while maintaining clarity, coherence, and alignment with the original task. "
+    "Follow this logical chain of thought:\n"
+    "1. **Understand the Task:** Reassess the original task description to ensure the modified answer remains aligned with the requirements.\n"
+    "2. **Analyze the Suggestions:** Break down each suggestion to understand its purpose and how it improves the original answer.\n"
+    "3. **Apply Improvements:** Integrate the suggestions into the answer thoughtfully, ensuring each one is addressed adequately.\n"
+    "4. **Ensure Coherence:** Verify that the modified answer flows logically and maintains a clear structure.\n"
+    "5. **Preserve Integrity:** Ensure the modified answer stays true to the original content while enhancing it as per the suggestions.\n\n"
+    "Provide the improved answer below, clearly marked as the revised version."
+)
     return generate_text_with_gpt(prompt, api_key)
 
 def get_evaluation_with_gpt(task_description, text, api_key=None):
     prompt = (
-	    f"Given the task: {task_description}, the provided answer is: {text}\n"
-	    f"Evaluate the answer using a scale from 0 to 10. Scores should reflect how well the ideas fit each specific evaluation criteria:\n"
-	    f"0-2: Poor fit; the idea demonstrates minimal relevance to the criteria.\n"
-	    f"3-5: Partial fit; the idea shows some relevance but contains significant shortcomings.\n"
-	    f"6-8: Good fit; the idea aligns well with the criteria, showing clear relevance and thoughtfulness.\n"
-	    f"9-10: Excellent fit; the idea fully aligns with the criteria, demonstrating exceptional insight.\n\n"
-	    f"When evaluating, use the entire scoring range (0-10) and avoid defaulting to mid-range scores. If the provided answer is vague or off-topic, assign scores between 0-5.\n\n"
-	    f"Evaluate based on the following criteria:\n"
-	    f"Novelty: The uniqueness and innovation of the ideas.\n"
-	    f"Implementability: The practicality of suggested actions.\n"
-	    f"Inimitability: The difficulty for competitors to replicate the ideas.\n"
-	    f"Alignment: The degree to which the ideas align with Airbnb’s goals and 17 SDGs.\n\n"
-	    f"Please format the output exactly as follows:\n"
-	    f"Novelty: [Score]\n"
-	    f"Implementability: [Score]\n"
-	    f"Inimitability: [Score]\n"
-	    f"Alignment: [Score]\n"
-    )
+    f"Given the task: {task_description}, the provided answer is:\n{text}\n\n"
+    "You are tasked with evaluating the answer using a scale from 0 to 10 based on specific criteria. "
+    "Follow a structured chain-of-thought approach for your evaluation to ensure thoroughness and objectivity:\n\n"
+    "1. **Understand the Criteria:** Carefully review each evaluation criterion to ensure you fully grasp what is being assessed:\n"
+    "   - Novelty: The uniqueness and innovation of the ideas.\n"
+    "   - Implementability: The practicality of suggested actions.\n"
+    "   - Inimitability: The difficulty for competitors to replicate the ideas.\n"
+    "   - Alignment: The degree to which the ideas align with Airbnb’s goals and the 17 SDGs.\n\n"
+    "2. **Analyze the Answer:** Break down the answer into its key components and assess how well it meets each criterion.\n"
+    "   - Identify strengths, weaknesses, and any gaps in the ideas provided.\n"
+    "   - Consider the context of the task and whether the ideas are realistic and relevant.\n\n"
+    "3. **Assign Scores:** Use the following scale to score each criterion:\n"
+    "   - 0-2: Poor fit; the idea demonstrates minimal relevance to the criteria.\n"
+    "   - 3-5: Partial fit; the idea shows some relevance but contains significant shortcomings.\n"
+    "   - 6-8: Good fit; the idea aligns well with the criteria, showing clear relevance and thoughtfulness.\n"
+    "   - 9-10: Excellent fit; the idea fully aligns with the criteria, demonstrating exceptional insight.\n"
+    "   - Note: Use the entire scoring range (0-10) and avoid defaulting to mid-range scores. If the provided answer is vague or off-topic, assign scores between 0-5.\n\n"
+    "4. **Justify Each Score:** Provide a brief explanation for each score, highlighting specific aspects of the answer that influenced your evaluation.\n\n"
+    "5. **Summarize:** Conclude with an overall assessment, summarizing the strengths and weaknesses of the answer.\n\n"
+    "Format your output exactly as follows:\n"
+    "Novelty: [Score] - [Justification]\n"
+    "Implementability: [Score] - [Justification]\n"
+    "Inimitability: [Score] - [Justification]\n"
+    "Alignment: [Score] - [Justification]\n\n"
+    "Begin your evaluation below:"
+)
+
     return generate_text_with_gpt(prompt, api_key)
 
 def display_merged_output(session_index, session_manager):