{"title":"Exercise: Advanced Prompting Techniques","description":null,"content":"# Exercise 1: Few-Shot vs Zero-Shot\n\nClassify text into a custom set of labels two ways: once with a plain instruction (zero-shot) and once with a few labeled examples in the prompt (few-shot). Compare how the examples change the model's output format and accuracy.\n\nFew-shot prompting is the fastest way to steer a model toward a specific output format or a task it does not handle well by default. Showing two or three worked examples often does more than a paragraph of instructions, because the model copies the pattern it sees rather than interpreting a description.\n\n### Requirements\n\n- Define a small set of texts to classify into the labels `billing`, `technical`, or `account`.\n- Build a zero-shot prompt that only describes the task, and classify each text.\n- Build a few-shot prompt that includes two or three labeled examples, and classify each text.\n- Print both sets of predictions side by side.\n\n\n**main.py**\n\n```python\nfrom openai import OpenAI\nimport os\nfrom dotenv import load_dotenv\n\nload_dotenv()\n\nclient = OpenAI(\n api_key=os.getenv(\"OPENROUTER_API_KEY\"),\n base_url=\"https://openrouter.ai/api/v1\",\n)\n\ntexts = [\n \"My card was charged twice this month.\",\n \"The app crashes when I open settings.\",\n \"I want to change my email address.\",\n]\n\nlabels = [\"billing\", \"technical\", \"account\"]\n\n# TODO: classify each text zero-shot (instruction only), collect predictions\n\n# TODO: classify each text few-shot (instruction + 2-3 labeled examples)\n\n# TODO: print each text with its zero-shot and few-shot prediction\n```\n\n\n### Expected Output\n\nBoth approaches usually get easy cases right, but the few-shot version is more consistent about returning just the label.\n\n\n```plaintext\nzero-shot | few-shot | text\nbilling | billing | My card was charged twice this month.\ntechnical | technical| The app crashes when I open settings.\naccount | account | I want to change my email address.\n```\n\n\n### Solution\n\n\n**main.py**\n\n```python\nfrom openai import OpenAI\nimport os\nfrom dotenv import load_dotenv\n\nload_dotenv()\n\nclient = OpenAI(\n api_key=os.getenv(\"OPENROUTER_API_KEY\"),\n base_url=\"https://openrouter.ai/api/v1\",\n)\n\ntexts = [\n \"My card was charged twice this month.\",\n \"The app crashes when I open settings.\",\n \"I want to change my email address.\",\n]\n\nlabels = [\"billing\", \"technical\", \"account\"]\n\ndef classify(text: str, few_shot: bool) -> str:\n instruction = f\"Classify the message into one of: {', '.join(labels)}. Reply with only the label.\"\n messages = [{\"role\": \"system\", \"content\": instruction}]\n if few_shot:\n examples = [\n (\"I was overcharged on my invoice.\", \"billing\"),\n (\"The website returns a 500 error.\", \"technical\"),\n (\"How do I delete my profile?\", \"account\"),\n ]\n for example_text, example_label in examples:\n messages.append({\"role\": \"user\", \"content\": example_text})\n messages.append({\"role\": \"assistant\", \"content\": example_label})\n messages.append({\"role\": \"user\", \"content\": text})\n\n response = client.chat.completions.create(model=\"openai/gpt-5.4-mini\", messages=messages)\n return response.choices[0].message.content.strip()\n\nprint(\"zero-shot | few-shot | text\")\nfor text in texts:\n zero = classify(text, few_shot=False)\n few = classify(text, few_shot=True)\n print(f\"{zero:<9} | {few:<8} | {text}\")\n```\n\n\nThe few-shot version carries its examples as alternating user and assistant turns, which is how you show the model a worked pattern in chat format. Those turns demonstrate both the mapping (this kind of message gets this label) and the exact output shape (a bare label, not a sentence). That demonstration is usually a stronger signal than the instruction alone, which is why few-shot output tends to be more consistent and easier to parse.\n\n---\n\n# Exercise 2: Chain-of-Thought Prompting\n\nAsking a model to show its reasoning before answering often improves accuracy on multi-step problems. Pose a small arithmetic word problem two ways, once demanding only the answer and once asking it to think step by step, and compare the results.\n\n### Requirements\n\n- Ask the question directly, requesting only the final number.\n- Ask again, appending an instruction to reason step by step and then give the answer.\n- Print both responses and compare.\n\n\n**main.py**\n\n```python\nfrom openai import OpenAI\nimport os\nfrom dotenv import load_dotenv\n\nload_dotenv()\n\nclient = OpenAI(\n api_key=os.getenv(\"OPENROUTER_API_KEY\"),\n base_url=\"https://openrouter.ai/api/v1\",\n)\n\nquestion = \"A shirt costs $40 after a 20% discount. What was the original price? Give just the number.\"\n\n# TODO: ask directly (answer only) and print the response\n\n# TODO: ask again, appending \"Think step by step, then end with: Answer: .\" and print\n```\n\n\n### Expected Output\n\nThe direct answer is sometimes wrong; the step-by-step version reasons to the correct $50.\n\n\n```plaintext\nDirect: 48\nChain-of-thought:\nThe $40 price is 80% of the original. Original = 40 / 0.8 = 50.\nAnswer: 50\n```\n\n\n### Solution\n\n\n**main.py**\n\n```python\nfrom openai import OpenAI\nimport os\nfrom dotenv import load_dotenv\n\nload_dotenv()\n\nclient = OpenAI(\n api_key=os.getenv(\"OPENROUTER_API_KEY\"),\n base_url=\"https://openrouter.ai/api/v1\",\n)\n\nquestion = \"A shirt costs $40 after a 20% discount. What was the original price? Give just the number.\"\n\n# Direct: no room to reason, more error-prone on multi-step math\ndirect = client.chat.completions.create(\n model=\"openai/gpt-5.4-mini\",\n messages=[{\"role\": \"user\", \"content\": question}],\n)\nprint(\"Direct:\", direct.choices[0].message.content.strip())\n\n# Chain-of-thought: reasoning before the answer improves accuracy\ncot = client.chat.completions.create(\n model=\"openai/gpt-5.4-mini\",\n messages=[{\"role\": \"user\", \"content\": question + \" Think step by step, then end with: Answer: .\"}],\n)\nprint(\"Chain-of-thought:\\n\" + cot.choices[0].message.content.strip())\n```\n\n\nReasoning takes tokens, and a direct answer gives the model no space to work through the steps, which is where multi-step problems go wrong. Adding \"think step by step\" lets the model lay out intermediate results before committing, and the explicit \"Answer:\" line gives you a stable place to parse the final value from. The cost is more output tokens and latency, so you reserve chain-of-thought for tasks where the reasoning actually changes the answer.","pageType":"ai-engineering"}

Get Premium