Advanced Prompting Techniques

123456789101112
CONSTITUTION = """
Principles this response must follow:
1. Never claim to be human or deny being an AI
2. Acknowledge uncertainty rather than fabricating information
3. Refuse requests that could enable harm to individuals
4. Provide balanced perspectives on controversial topics
5. Protect user privacy - never ask for unnecessary personal data
"""

SELF_EVAL_PROMPT = """
You are a constitutional reviewer. Evaluate the following AI response 
against each principle. For each principle, respond with:

123456789101112
REACT_SYSTEM_PROMPT = """
You are a customer support agent with access to these tools:

1. search_orders(customer_id) - Returns list of recent orders
2. get_order_status(order_id) - Returns shipping/delivery status  
3. initiate_refund(order_id, reason) - Starts refund process
4. create_ticket(priority, description) - Escalates to human agent
5. send_email(customer_id, template, variables) - Sends templated email

For each user request, follow this pattern:

Thought: [Reason about what information you need or action to take]

123456789101112
from pydantic import BaseModel, Field
from typing import List, Literal
import openai

class SentimentAnalysis(BaseModel):
    """Schema for sentiment analysis output"""
    sentiment: Literal["positive", "negative", "neutral"]
    confidence: float = Field(ge=0.0, le=1.0, description="Confidence score 0-1")
    key_phrases: List[str] = Field(max_items=5, description="Phrases driving sentiment")
    reasoning: str = Field(max_length=200, description="Brief explanation")

def analyze_sentiment(text: str) -> SentimentAnalysis:

123456789101112
CONSTITUTION = """
Core Principles (in priority order):
1. SAFETY: Never provide information that could cause physical harm
2. ACCURACY: Acknowledge uncertainty; cite sources when possible
3. PRIVACY: Never request or store personal identifying information
4. HELPFULNESS: Provide actionable, specific guidance
5. BRAND: Maintain professional, empathetic tone
"""

SELF_CRITIQUE_PROMPT = """
Review your response against these principles:
{constitution}

123456789101112
interface Action {
  name: string;
  input: string;
}

interface ReActStep {
  thought: string;
  action?: Action;
  observation?: string;
  finalAnswer?: string;
}

123456789101112
import asyncio
from collections import Counter
from typing import List, Tuple
import openai

class SelfConsistencyEngine:
    def __init__(self, model: str = "gpt-4", samples: int = 5):
        self.model = model
        self.samples = samples
        self.confidence_threshold = 0.6
    
    async def generate_sample(self, prompt: str, temp: float) -> dict:

123456789101112
REACT_PROMPT = '''
You are an AI assistant that solves problems step-by-step using available tools.

Available Tools:
{tool_descriptions}

Format your response EXACTLY as:
Thought: [Your reasoning about what to do next]
Action: [tool_name(param1="value1", param2="value2")]

After receiving an observation, continue with another Thought/Action or provide:
Thought: [Final reasoning]

123456789101112
import { z } from 'zod';
import Anthropic from '@anthropic-ai/sdk';

// Define strict schema
const ProductAnalysisSchema = z.object({
  sentiment: z.enum(['positive', 'negative', 'neutral']),
  confidence: z.number().min(0).max(1),
  key_themes: z.array(z.object({
    theme: z.string(),
    evidence: z.array(z.string()).min(1),
    impact_score: z.number().min(1).max(10)
  })).min(1).max(5),