3.1 Prompt_Engineering.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "039ae489-7705-497c-98a5-1d1c087e31ad",
   "metadata": {},
   "source": [
    "# Prompt Engineering\n",
    "Language Models (LLMs) are designed to respond to prompts, representing a novel method of programming computers. Crafting effective prompts is both an art and a science, aimed at optimizing a model's performance for specific tasks. By presenting LLMs with structured prompts resembling complete documents, we can leverage their ability to generate responses tailored to our needs.\n",
    "\n",
    "Interacting with LLMs is straightforward through their websites like ChatGPT, Bard, etc., although these platforms have limitations. A more versatile approach involves programmatically interacting with LLMs using languages like Python, JavaScript, etc. This method unlocks a new paradigm in software development, enabling the creation of AI applications through prompt engineering.\n",
    "\n",
    "Consider the use case of creating an automatic grader for a high school history class. Traditionally, developers would need to devise logic to handle various answer variations, possibly requiring a list of all potential correct answers and employing exact or fuzzy string matching algorithms. In the new paradigm, developers can offload the logic to the LLM through prompt engineering.\n",
    "\n",
    "Effective prompt engineering involves several key strategies. First, prompts should be descriptive and provide examples to guide the LLM's response. Structured text can help organize the prompt's content, while a chain of thoughts can lead the LLM to generate more coherent responses. Assigning a chatbot persona to the LLM, such as a role or expertise, can further enhance its responses. Another approach is to let the LLM ask questions, allowing developers to extract information and generate more comprehensive responses. Finally, developers should regularly review and refine past interactions to improve the LLM's performance over time."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5134c2b2-9d33-4f17-9518-8d243c83d577",
   "metadata": {},
   "source": [
    "## Prompting Gemini-Pro"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "77b99fed-cd2c-46a1-a923-666388181ed6",
   "metadata": {},
   "outputs": [],
   "source": [
    "!pip install -q -U google-generativeai\n",
    "import google.generativeai as genai"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "9a2370ec-10f2-4c09-b8ad-fdb98a9f26bd",
   "metadata": {},
   "outputs": [],
   "source": [
    "GOOGLE_API_KEY = \"YOUR_API_KEY\"\n",
    "genai.configure(api_key=GOOGLE_API_KEY)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "13f4670d-0e02-42b9-acb4-58e1172a4fcd",
   "metadata": {},
   "outputs": [],
   "source": [
    "model = genai.GenerativeModel('gemini-pro')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "eeec51a8-0c8a-4a2c-ad06-e3e93c6dfde8",
   "metadata": {},
   "outputs": [],
   "source": [
    "from IPython.display import display\n",
    "from IPython.display import Markdown\n",
    "import textwrap\n",
    "\n",
    "def response(text):\n",
    "    res = model.generate_content(text).text\n",
    "    text = res.replace('•', '  *')\n",
    "    return Markdown(textwrap.indent(text, '> ', predicate=lambda _: True))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "e06405f5-7f15-4e99-8461-74084e060924",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/markdown": [
       "> I am well, thank you for asking. How are you doing today?"
      ],
      "text/plain": [
       "<IPython.core.display.Markdown object>"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "response('Hi how are you?')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "8d8aa4b7-7e5d-40ec-814e-2675c297109d",
   "metadata": {},
   "outputs": [],
   "source": [
    "prompt = \"\"\"I am planning a road trip across the United States.\n",
    "Can you suggest the best route to take and must-visit landmarks along the way?\"\"\""
   ]
  },
  {
   "cell_type": "markdown",
   "id": "dec7996b-f296-4256-be05-29f44d47f5d6",
   "metadata": {},
   "source": [
    "The above prompt has two parts, **context** `(I am planning a road trip across the United States.)` and **task** `(Can you suggest the best route to take and must-visit landmarks along the way?)`. In this prompt the task is of more importance, because if we just pass the **task** we will still get some useful response but by passing only the **context** we won't get anything."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "a3da4e25-6d72-45a5-9726-5a81b2a520fa",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/markdown": [
       "> **Optimal Route:**\n",
       "> \n",
       "> **East Coast to West Coast:**\n",
       "> \n",
       "> * Start in New York City\n",
       "> * Drive across Pennsylvania and Ohio\n",
       "> * Visit Niagara Falls in New York\n",
       "> * Explore St. Louis, Missouri\n",
       "> * Cross Nebraska, Wyoming, and Utah\n",
       "> * Enter California at Lake Tahoe\n",
       "> * Arrive in San Francisco\n",
       "> \n",
       "> **Alternate Route (South-Central):**\n",
       "> \n",
       "> * Begin in New Orleans, Louisiana\n",
       "> * Travel through Mississippi, Alabama, and Georgia\n",
       "> * Visit Graceland in Memphis, Tennessee\n",
       "> * Cross Arkansas, Oklahoma, and Texas\n",
       "> * Enter New Mexico and visit Carlsbad Caverns National Park\n",
       "> * Cross Arizona and California\n",
       "> * Arrive in Los Angeles\n",
       "> \n",
       "> **Must-Visit Landmarks:**\n",
       "> \n",
       "> **East Coast:**\n",
       "> \n",
       "> * **Statue of Liberty and Ellis Island (New York City)**\n",
       "> * **Lincoln Memorial and Washington Monument (Washington D.C.)**\n",
       "> * **Liberty Bell and Independence Hall (Philadelphia, Pennsylvania)**\n",
       "> * **Baltimore Inner Harbor (Maryland)**\n",
       "> \n",
       "> **Midwest:**\n",
       "> \n",
       "> * **Niagara Falls (New York)**\n",
       "> * **Mount Rushmore (South Dakota)**\n",
       "> * **Yellowstone National Park (Wyoming and Montana)**\n",
       "> * **Badlands National Park (South Dakota)**\n",
       "> \n",
       "> **Western States:**\n",
       "> \n",
       "> * **Grand Canyon National Park (Arizona)**\n",
       "> * **Lake Tahoe (Nevada and California)**\n",
       "> * **Yosemite National Park (California)**\n",
       "> * **Sequoia National Park (California)**\n",
       "> \n",
       "> **Additional Tips:**\n",
       "> \n",
       "> * Allow for plenty of time for driving and sightseeing.\n",
       "> * Book accommodations and rental cars in advance, especially during peak season.\n",
       "> * Research and plan your route carefully to avoid unnecessary detours.\n",
       "> * Consider visiting national parks and historical sites for unique experiences.\n",
       "> * Be prepared for varying weather conditions.\n",
       "> * Pack plenty of snacks and water for long stretches of road.\n",
       "> * Take breaks and enjoy the scenery along the way."
      ],
      "text/plain": [
       "<IPython.core.display.Markdown object>"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "response(prompt)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "292fbbb7-912f-43ff-a7da-098df276a393",
   "metadata": {},
   "source": [
    "Try to pass context only and task only"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "b9d37666-da1f-4f50-b2d6-25ad07464bb3",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/markdown": [
       "> **Planning a Cross-Country Road Trip Across the United States**\n",
       "> \n",
       "> **Step 1: Determine Your Route**\n",
       "> \n",
       "> * Consider your interests, time constraints, and budget.\n",
       "> * Plan a loop or one-way trip, visiting iconic landmarks, national parks, and major cities.\n",
       "> * Use online tools like Google Maps or Roadtrippers to map your route.\n",
       "> \n",
       "> **Step 2: Set a Budget**\n",
       "> \n",
       "> * Estimate expenses for gas, accommodation, food, entertainment, and unforeseen circumstances.\n",
       "> * Research accommodation options along your route, including hotels, motels, and campgrounds.\n",
       "> * Consider purchasing a National Parks Pass if you plan to visit multiple parks.\n",
       "> \n",
       "> **Step 3: Prepare Your Vehicle**\n",
       "> \n",
       "> * Ensure your car is in good mechanical condition and up-to-date on maintenance.\n",
       "> * Pack a basic toolkit, spare tire, and jumper cables.\n",
       "> * Consider purchasing roadside assistance insurance for peace of mind.\n",
       "> \n",
       "> **Step 4: Pack Essentials**\n",
       "> \n",
       "> * Clothing and toiletries for all weather conditions\n",
       "> * First-aid kit and any necessary medications\n",
       "> * Entertainment items like books, music, or games\n",
       "> * Snacks and non-perishable food items\n",
       "> * Navigation system or smartphone with GPS\n",
       "> * Phone charger and portable battery pack\n",
       "> \n",
       "> **Step 5: Plan Your Stops**\n",
       "> \n",
       "> * Research national parks, monuments, and historical sites along your route.\n",
       "> * Book accommodations in advance, especially during peak season.\n",
       "> * Allow for unplanned stops at roadside attractions or scenic views.\n",
       "> \n",
       "> **Step 6: Set a Timeline**\n",
       "> \n",
       "> * Plan a realistic daily driving distance to avoid fatigue.\n",
       "> * Factor in time for rest stops, sightseeing, and potential delays.\n",
       "> * Allow for flexibility to adjust your schedule as needed.\n",
       "> \n",
       "> **Step 7: Stay Safe and Legal**\n",
       "> \n",
       "> * Always obey traffic laws and speed limits.\n",
       "> * Designate a responsible driver and take regular breaks to avoid drowsiness.\n",
       "> * Inform someone of your itinerary and check in with them periodically.\n",
       "> \n",
       "> **Additional Tips:**\n",
       "> \n",
       "> * Start your trip during shoulder season for smaller crowds and lower costs.\n",
       "> * Pack light to maximize fuel efficiency and minimize luggage handling.\n",
       "> * Learn some basic car maintenance skills in case of emergencies.\n",
       "> * Take advantage of rest areas and roadside parks for free overnight camping in some areas.\n",
       "> * Consider using a ride-sharing app or public transportation for short distances in major cities.\n",
       "> * Embrace the adventure and enjoy the journey!"
      ],
      "text/plain": [
       "<IPython.core.display.Markdown object>"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "response('I am planning a road trip across the United States.')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "becf29d7-757b-4379-86d7-4f9bbc8f970f",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/markdown": [
       "> **East Coast Adventure**\n",
       "> \n",
       "> * **Start:** New York City, NY\n",
       "> * **End:** Miami, FL\n",
       "> \n",
       "> **Must-Visit Landmarks:**\n",
       "> \n",
       "> * **New York City:** Statue of Liberty, Times Square, Empire State Building\n",
       "> * **Philadelphia, PA:** Liberty Bell, Independence Hall\n",
       "> * **Washington, D.C.:** Lincoln Memorial, Washington Monument, Smithsonian Museums\n",
       "> * **Charleston, SC:** Charleston Historic District, Battery Park\n",
       "> * **Savannah, GA:** Forsyth Park, Bonaventure Cemetery\n",
       "> * **St. Augustine, FL:** Castillo de San Marcos, St. Augustine Historical District\n",
       "> * **Miami, FL:** South Beach, Wynwood Walls, Everglades National Park\n",
       "> \n",
       "> **West Coast Wonders**\n",
       "> \n",
       "> * **Start:** Los Angeles, CA\n",
       "> * **End:** Seattle, WA\n",
       "> \n",
       "> **Must-Visit Landmarks:**\n",
       "> \n",
       "> * **Los Angeles:** Hollywood Walk of Fame, Griffith Observatory, Santa Monica Pier\n",
       "> * **San Francisco, CA:** Golden Gate Bridge, Alcatraz Island, Fisherman's Wharf\n",
       "> * **Yosemite National Park:** Half Dome, El Capitan, Yosemite Falls\n",
       "> * **Portland, OR:** International Rose Test Garden, Powell's City of Books\n",
       "> * **Cannon Beach, OR:** Haystack Rock\n",
       "> * **Seattle, WA:** Space Needle, Pike Place Market, Mount Rainier National Park\n",
       "> \n",
       "> **Route 66 Odyssey**\n",
       "> \n",
       "> * **Start:** Chicago, IL\n",
       "> * **End:** Santa Monica, CA\n",
       "> \n",
       "> **Must-Visit Landmarks:**\n",
       "> \n",
       "> * **Chicago:** Willis Tower, Navy Pier, Art Institute of Chicago\n",
       "> * **St. Louis, MO:** Gateway Arch, City Museum\n",
       "> * **Oklahoma City, OK:** National Cowboy & Western Heritage Museum\n",
       "> * **Amarillo, TX:** Cadillac Ranch\n",
       "> * **Santa Fe, NM:** Georgia O'Keeffe Museum, Palace of the Governors\n",
       "> * **Williams, AZ:** Grand Canyon Railway\n",
       "> * **Los Angeles:** Santa Monica Pier, Venice Beach\n",
       "> \n",
       "> **National Parks Explorer**\n",
       "> \n",
       "> * **Start:** Yellowstone National Park, WY\n",
       "> * **End:** Grand Canyon National Park, AZ\n",
       "> \n",
       "> **Must-Visit Landmarks:**\n",
       "> \n",
       "> * **Yellowstone National Park:** Old Faithful, Grand Prismatic Spring, Mammoth Hot Springs\n",
       "> * **Grand Teton National Park:** Grand Teton, Jenny Lake\n",
       "> * **Rocky Mountain National Park:** Trail Ridge Road, Estes Park\n",
       "> * **Arches National Park:** Delicate Arch, Devil's Garden\n",
       "> * **Canyonlands National Park:** Island in the Sky, Horseshoe Canyon\n",
       "> * **Zion National Park:** Angels Landing, The Narrows\n",
       "> * **Bryce Canyon National Park:** Bryce Canyon Amphitheater, Sunset Point\n",
       "> * **Grand Canyon National Park:** Grand Canyon, South Rim, North Rim\n",
       "> \n",
       "> **Music Lover's Pilgrimage**\n",
       "> \n",
       "> * **Start:** Memphis, TN\n",
       "> * **End:** Seattle, WA\n",
       "> \n",
       "> **Must-Visit Landmarks:**\n",
       "> \n",
       "> * **Memphis:** Graceland, Sun Studio, Beale Street\n",
       "> * **Nashville, TN:** Country Music Hall of Fame, Ryman Auditorium, Tootsie's Orchid Lounge\n",
       "> * **Austin, TX:** Sixth Street, The Continental Club, Austin City Limits Live\n",
       "> * **New Orleans, LA:** Preservation Hall, Frenchmen Street, New Orleans Jazz National Historical Park\n",
       "> * **Seattle:** Experience Music Project, EMP Museum, Jimi Hendrix Home\n"
      ],
      "text/plain": [
       "<IPython.core.display.Markdown object>"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "response('Can you suggest the best route to take and must-visit landmarks along the way in United States?')"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5dd9c4fb-e84e-44e0-8c24-ee5a83ff3e48",
   "metadata": {},
   "source": [
    "As we can see by just passing the **task** we get similar response as we get when passing the entire prompt. But the response we get while just passing the context is not useful for us."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "34d6104b-421a-48b0-89b1-c6160c4a9da8",
   "metadata": {},
   "source": [
    "Generally it is better to start the task with an **Action Verb (eg. Generate, Give, Write, Analyze, etc)**.<br> \n",
    "**Prompt :-** `\"I am planning a road trip across the United States. Suggesting the best route to take and must-visit landmarks along the way. I am looking for a road trip route that covers the United States. Provide a suggested route and must-visit landmarks for a road trip across the United States.\"`<br>\n",
    "This is a more brief way to express the above used prompt."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b11d82c4-2065-4493-bfd6-8a69e289f686",
   "metadata": {},
   "source": [
    "## Zero Shot Learning\n",
    "Zero-shot learning refers to the ability of the model to perform tasks or generate outputs without explicit training on those tasks or examples."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "e8b3290a-a08e-4933-bdcb-d010930c28d2",
   "metadata": {},
   "outputs": [],
   "source": [
    "zero_shot_prompt = \"\"\"Generate a short story about a unicorn that\n",
    "becomes friends with a group of aliens on a distant planet.\"\"\""
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ffe7bfc9-1b02-451f-90bb-4a6e65343d39",
   "metadata": {},
   "source": [
    "The model is expected to use its general understanding of language and storytelling to come up with a coherent and imaginative story based on the prompt."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "3f0aba64-29f4-4548-93f3-eb269c011112",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/markdown": [
       "> In the ethereal realm of Aurelia, where celestial wonders danced in harmony, there lived Celeste, a unicorn of iridescent beauty. Her ivory mane shimmered like a thousand stars, and her azure eyes sparkled with an otherworldly light.\n",
       "> \n",
       "> One fateful night, as Celeste grazed upon a meadow of moonlit flowers, a peculiar sound caught her attention. Looking up, she beheld a radiant sphere descending from the heavens. Its sleek exterior glimmered with iridescent colors, and strange music emanated from within.\n",
       "> \n",
       "> As the sphere touched down, a hatch opened, revealing a group of aliens. They were ethereal beings with iridescent skin, shimmering eyes, and antennae that reached towards the cosmos. Led by Lyra, a wise and enigmatic alien, they greeted Celeste with a chorus of curious clicks and whistles.\n",
       "> \n",
       "> Despite their differences, Celeste and the aliens felt an instant kinship. Lyra introduced her companions: Aethon, the swift and agile; Nyx, the mysterious and ethereal; and Alcor, the playful and curious. Together, they explored the wonders of Aurelia, sharing stories of their distant home and marveling at the beauty of Celeste's realm.\n",
       "> \n",
       "> Days turned into nights as the bond between Celeste and the aliens grew stronger. They rode through enchanted forests, their laughter echoing through the ethereal expanse. They swam in moonlit rivers, their combined brilliance illuminating the depths.\n",
       "> \n",
       "> However, their idyllic sanctuary was soon threatened by a dark force from the heavens. A celestial predator, known as the Shadow, pursued the aliens relentlessly, its jagged claws and glowing eyes casting a sinister shadow over Aurelia.\n",
       "> \n",
       "> Undeterred, Celeste and her alien friends devised a plan to vanquish the Shadow. Using her celestial magic and their advanced technology, they lured the predator into a trap within the heart of an ancient forest. With a collective burst of light and energy, they defeated the Shadow, restoring peace and tranquility to Aurelia.\n",
       "> \n",
       "> As the aliens prepared to return to their distant star, they bid farewell to Celeste with heavy hearts. Lyra promised that they would never forget their extraordinary friendship and adventures on Aurelia.\n",
       "> \n",
       "> And so, in the annals of celestial lore, the tale of Celeste, the unicorn, and her alien companions became a testament to the boundless power of friendship and the interconnectedness of all beings across the vast expanses of the cosmos."
      ],
      "text/plain": [
       "<IPython.core.display.Markdown object>"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "response(zero_shot_prompt)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3c54ceb8-4864-4130-b770-a40fa9ee4533",
   "metadata": {},
   "source": [
    "## One Shot Learning\n",
    "One-shot learning refers to the ability of the model to perform tasks or generate outputs with minimal examples or demonstrations. We can provide a single example or demonstration of the task you want the model to perform, and it can generalize from that example to generate outputs for similar tasks.This is particularly useful when you want the model to perform a task that it has not been explicitly trained on, but for which you can provide a simple example or description to guide its behavior."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "60d598f1-834f-4e34-a4f7-b2b67bde458c",
   "metadata": {},
   "outputs": [],
   "source": [
    "one_shot_prompt = \"\"\"Generate a short story about a detective solving a mysterious disappearance.\n",
    "Example Story: \"Detective Smith was known for his sharp wit and keen eye for detail. When Mrs.\n",
    "Johnson's prized necklace went missing, he was called in to solve the case...\"\n",
    "\"\"\""
   ]
  },
  {
   "cell_type": "markdown",
   "id": "bf7d145b-4c80-4ef4-84fa-bc0b22b58496",
   "metadata": {},
   "source": [
    "In this prompt we have provided the example in which way we want our output. The model is not explicitly trained on this type of data but it provides a structured answer by matching the  provided example."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "5d0f54cc-9019-44a9-9c40-a314f8aad4e1",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/markdown": [
       "> In the mist-shrouded town of Willow Creek, Detective Emily Carter had carved a reputation as an astute investigator. When the enigmatic recluse, Dr. Hawthorne, vanished without a trace, the enigmatic circumstances drew her into a perplexing labyrinth.\n",
       "> \n",
       "> The labyrinth began with Dr. Hawthorne's secluded mansion, where his lingering absence was a chilling testament to his disappearance. Emily's keen eye scrutinized every nook and cranny, seeking clues that could unravel the puzzle.\n",
       "> \n",
       "> Amidst the dust and forgotten memories, she stumbled upon an encrypted journal. Deciphering its arcane script, Emily discovered that Dr. Hawthorne had been delving into forbidden research that could alter the very fabric of reality.\n",
       "> \n",
       "> As the investigation deepened, Emily encountered cryptic messages and enigmatic individuals who seemed to hold fragments of the missing doctor's fate. Each encounter led her deeper into a web of intrigue that stretched beyond the boundaries of the known world.\n",
       "> \n",
       "> With each piece of evidence, Emily's determination grew, driving her to unearth the truth. She delved into hidden archives, unraveled coded transmissions, and navigated treacherous landscapes.\n",
       "> \n",
       "> Finally, after weeks of meticulous investigation, Emily confronted the orchestrator behind Dr. Hawthorne's disappearance. In a thrilling showdown, she exposed the sinister plot and rescued the missing scientist, bringing closure to the enigmatic case that had captivated Willow Creek."
      ],
      "text/plain": [
       "<IPython.core.display.Markdown object>"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "response(one_shot_prompt)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "aec0272e-028e-45a0-8636-9c234a0ee8f9",
   "metadata": {},
   "source": [
    "## Few Shot Learning\n",
    "In few-shot learning, the model is typically given a few examples (usually less than five) of the task it needs to perform, along with the desired outcome, and is then asked to generalize from these examples to perform the task on new examples.<br>\n",
    "\n",
    "\n",
    "In this prompt we have provided four examples with desired answer and ask the model to provide the answer for the fifth example."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "8e9d91dc-5254-45ee-b826-a4063ab26aa4",
   "metadata": {},
   "outputs": [],
   "source": [
    "few_shot_prompt = \"\"\"\n",
    "The odd numbers in this group add up to an even number: 4, 8, 9, 15, 12, 2, 1.\n",
    "A: The answer is False.\n",
    "The odd numbers in this group add up to an even number: 17,  10, 19, 4, 8, 12, 24.\n",
    "A: The answer is True.\n",
    "The odd numbers in this group add up to an even number: 16,  11, 14, 4, 8, 13, 24\n",
    "A: The answer is True.\n",
    "The odd numbers in this group add up to an even number: 17,  9, 10, 12, 13, 4, 2\n",
    "A: The answer is False.\n",
    "The odd numbers in this group add up to an even number: 15, 32, 5, 13, 82, 7, 1.\n",
    "A: \"\"\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "d0817bff-cb8a-4053-8302-61691cdc800d",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/markdown": [
       "> The answer is False."
      ],
      "text/plain": [
       "<IPython.core.display.Markdown object>"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "response(few_shot_prompt)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "500ad20d-fbbf-4c31-9ea4-9561e782567b",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.11.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}