Lesson 2 Lecture

What is Prompt Engineering?

Prompt engineering is the art and science of crafting effective instructions for AI systems like Deepseek. Think of it as learning how to "speak AI" - finding the right words and structures that help these advanced systems understand exactly what you want them to do.

Just like you might phrase a question differently when talking to a teacher versus a friend, prompt engineering involves adapting your language to communicate clearly with AI systems. It's about finding the perfect balance between being specific enough to get what you want, but open enough to let the AI's capabilities shine.

For example, instead of asking "Tell me about space," a well-engineered prompt might be: "Explain three fascinating discoveries about black holes from the last decade in simple terms that a high school student would understand."

Why Use Prompt Engineering?

Imagine having a super-smart friend who takes everything extremely literally. That's basically what Deepseek and other AI systems are like! Here's why prompt engineering matters:

1. Get better results - The quality of an AI's output directly depends on the quality of your input. Better prompts = better responses.
2. Save time - Instead of going back and forth with an AI system trying to get what you want, a well-crafted prompt can get you there in one go.
3. Unlock hidden capabilities - Many AI systems can do amazing things, but only if you know how to ask! Prompt engineering helps you tap into capabilities you might not know exist.
4. Gain a valuable skill - As Deepseek and other AI models become more integrated into everyday life, knowing how to effectively communicate with these systems is becoming as important as knowing how to use a smartphone.
5. Take control - Rather than accepting whatever an AI gives you, prompt engineering puts you in the driver's seat of the interaction.

Real-World Uses of Prompt Engineering

Prompt Engineering is used (or at least, can be used) in all sorts of real life situations. For example, try and come up with the prompts that you would use for at least 5 examples below. How different are the prompts? How did you have to structure some of them differently?

For School Projects

• Transform a vague homework assignment into specific questions that AI can help you explore
• Generate creative writing prompts or story ideas based on specific themes
• Create study guides that focus on exactly what you need to learn

For Creative Projects

• Design detailed image prompts for AI art generators to create exactly the visual you're imagining
• Craft character profiles that help AI generate consistent dialogue for stories
• Develop music creation prompts that specify genre, mood, and instruments

For Practical Tasks

• Write prompts that help AI break down complex problems into manageable steps
• Create templates for emails, essays, or other writing projects
• Generate code snippets by describing exactly what functionality you need

For Learning

• Design prompts that challenge AI to explain concepts at different levels of complexity
• Create scenarios where AI can simulate debates between different viewpoints
• Transform dense information into easily understandable formats

Tips for Better Prompt Engineering

1. Be specific - Include details about format, length, style, audience, and purpose
2. Use examples - Show the AI what good output looks like
3. Break it down - Complex tasks work better when broken into steps
4. Experiment - Try different approaches and see what works best
5. Iterate - Use the AI's response to refine your prompt and try again

Remember: Prompt engineering isn't about tricking AI or finding hacks - it's about clear communication and collaboration with these powerful tools!

Running Deepseek Locally with Langchain and Ollama

Want to run Deepseek on your own computer? You can do it with Langchain and Ollama! Here's how:

Step 1: Install the Required Package

First, open your terminal or command prompt and install the Langchain-Ollama package:

pip install -U langchain-ollama

Step 2: Write the Python Code

Create a new Python file and add this code:

from langchain_ollama import OllamaLLM
from langchain_core.prompts import ChatPromptTemplate

template = """question: {question}
answer: Let's think step by step."""

prompt = ChatPromptTemplate.from_template(template)

model = OllamaLLM(model="deepseek-r1:7b")

chain = prompt | model

chain.invoke({"question": "What is the capital of France?"})

Step 3: Run and See the Results

When you run this code, Deepseek will process your question and produce output like this:

<think>
Okay, so I need to figure out what the capital of France is. Hmm, I've heard a lot about France in history and culture class, but I'm not entirely sure which city it is. Let me try to remember any information from my classes or maybe something I've read before.

I know that Paris is often mentioned as a major city in France. I think it's the capital because when I watch movies set in Europe, Paris usually pops up as a significant location for historical events. But wait, I'm not 100% certain if it's definitely Paris or another city like Lyon or Marseille. Maybe I should break this down.

First, let me recall what a capital city is. The capital of a country is typically the seat of government, including the executive branch (the president), and sometimes the legislative body (parliament). So France must have its government quarters there. But where exactly?

I remember learning that Paris was built on top of an ancient Roman camp called the Palais Royal, which dates back to Roman times. That might indicate that it's been a central hub for a long time. Also, I think Napoleon was born in France and later became Emperor of the French, so maybe his influence made Paris more important.

Wait, but I also heard about Lyon being a major city there. Is it possible that Lyon is the capital? Or perhaps Marseille or another city? I'm trying to remember if any of my teachers mentioned it specifically as Paris. I think it was definitely Paris because it's often referred to as "La Capital de la France." Yeah, that phrase rings a bell.

Let me also consider geography. France is a large country with several regions, but the largest city by area is probably Paris, right? It covers a significant portion of the mainland and has a massive population. That makes sense because if it's the capital, it should be one of the larger cities in terms of urban area.

Another point is that many international events take place in Paris, like the Eiffel Tower, which is super famous. If they built such iconic structures there, it seems logical that it's a major hub for politics and culture as well.

Wait, but isn't Lyon known for its universities? I think the ecole Polytechnique was originally in Paris, but maybe later moved to Lyon. No, actually, I'm confusing it with other institutions. Maybe that doesn't matter much. The main point is that Paris is where the government and key institutions are located.

I also recall that the French flag has a white arc across a blue background, and while this varies by region, it's often associated with Paris because of its prominence in the national identity. So, tying all these thoughts together, it seems like Paris is indeed the capital.

But just to be thorough, maybe I should think about any other capitals I know. For example, Brazil's capital is Braselia, which is a bit far from Rio de Janeiro. Italy's capital is Rome, so France being Paris makes sense because they are both in Europe and have historical significance.

I don't remember any other major European countries where the city isn't as prominent as Paris. So putting it all together, I think I can confidently say that the capital of France is Paris.
</think>

The capital of France is Paris.

What's Happening Here?

1. The code sets up a template that tells Deepseek to think step by step
2. We create a prompt template that can take variables (like our question)
3. We connect to the Deepseek model (the 7B parameter version)
4. We create a simple chain: prompt model
5. We run the chain with our question

Notice how Deepseek shows its thinking process with the