Learning by Experimentation

Whenever I start learning something new, I’ve noticed one thing consistently helps: experimentation.
Without understanding the foundational concepts, complex terminology and fast-moving AI trends rarely make sense.

My approach is simple:

1. Pick a platform

2. Read the official documentation

3. Experiment with small, focused examples

4. Learn how each concept behaves in practice

For this journey, I chose Claude and started directly from its documentation.

---

Starting Simple with Claude Haiku

To keep things lightweight and focused, I began with the Claude Haiku model.
It’s fast, cost-efficient, and ideal for understanding how inference parameters work without too much noise.

I used the claude-3-haiku-20240307 model to build a very simple chatbot that interacts with the Claude API.

Key Parameters to Understand

Before diving into the code, there are a few important parameters worth understanding:

• max_tokens
  Sets the maximum length of the model’s output (measured in tokens).

• temperature
  Controls how creative or random the model’s responses are.Ranges from 0.0 to 1.0

• Top-K Sampling
  Restricts the model to only consider the K most likely tokens at each step.

• Top-P (Nucleus Sampling)
  Allows the model to consider the smallest group of tokens whose combined probability is at least P.

In simple terms:

• Bigger max_tokens → longer responses

• Smaller max_tokens → concise responses, possibly cut off

• Lower temperature → predictable, factual output

• Higher temperature → more expressive and varied language

A Simple Claude Chatbot Example

Below is a minimal chatbot loop that sends user input to Claude and prints the response:

import os
from anthropic import Anthropic

client = Anthropic()

while True:
    user_input = input("You: ")
    if user_input.lower() in ["exit", "quit"]:
        break

    response = client.messages.create(
        model="claude-3-haiku-20240307",
        max_tokens=100,
        messages=[
            {"role": "user", "content": user_input}
        ],
        temperature=0.0
    )

    print("Claude:", response.content[0].text)

---

Observing Temperature in Action

To understand how temperature affects output, I used a simple prompt:

Input

Describe India

---

Output with temperature = 0.0

India is a vast and diverse country located in South Asia. It is the second-most populous country in the world, with over 1.3 billion people, and is known for its rich cultural heritage, vibrant traditions, and stunning natural landscapes.
Here are some key aspects of India:
1. Geography: India is a peninsula, bordered by the Indian Ocean, the Arabian Sea, and the Bay of Bengal...

---

Output with temperature = 1.0

Here's a brief overview of India:

Geography:
- India is a large country located in South Asia, covering an area of over 3.2 million square km.
- It is bordered by the Arabian Sea, the Bay of Bengal, and the Indian Ocean.
- The geography is diverse, ranging from the Himalayan mountains in the north to tropical rainforests in the south.

---

Key Learning

Even though the information remains correct, changing the temperature alters:

• sentence structure

• verbosity

• tone and flow

This small experiment made it clear that parameters don’t change knowledge — they change expression.

---

Final Thought

Reading documentation gives you the theory, but running small experiments builds intuition.
Once these fundamentals are clear, more advanced concepts and newer AI capabilities start to make much more sense.

This is just the beginning of my learning journey—and I plan to keep experimenting as I go.