Generating new content

Introduction to Claude Models

Nikhil Rangarajan

Data Scientist

Creative content generation

Beyond text transformation to original creation
- Stories

An open book

Creative content generation

Beyond text transformation to original creation
- Stories
- Marketing copy

An open book and a webpage showing marketing copy

Creative content generation

Beyond text transformation to original creation
- Stories
- Marketing copy
- Code

An open book, a webpage showing marketing copy and a code snippet

Creative content generation

Beyond text transformation to original creation
- Stories
- Marketing copy
- Code
- Emails

An open book, a webpage showing marketing copy, a code snippet and an email

Creative content generation

Beyond text transformation to original creation
- Stories
- Marketing copy
- Code
- Emails
- Scripts

An open book, a webpage showing marketing copy, a code snippet, an email, and a script

Crafting effective generation prompts

Specificity
Vague prompt: Generic results
Detailed prompt: Targeted content

# Vague prompt
prompt = "Write a product description"

# Specific prompt
prompt = """Write a 150-word product 
description for a wireless bluetooth 
speaker targeting college students. Use 
an energetic, casual tone and highlight 
portability, battery life, and sound 
quality."""

Crafting effective generation prompts

Most common Claude use case: code
Specific prompts drive targeted code outputs
- Programming language
- Specific functionality
- Error handling requirements
- Expected outputs

# Vague prompt
prompt = "Write some code"

# Specific prompt  
prompt = """Write a Python function 
that validates email addresses 
using regex. 
Include error handling for invalid 
inputs and return both boolean result 
and descriptive error message."""

Understanding temperature

Temperature range: 0.0 (deterministic) to 1.0 (highly creative)
- Low temperature (0.1-0.3): Consistent, predictable outputs
- Medium temperature (0.4-0.7): Balanced creativity and coherence
- High temperature (0.8-1.0): Maximum creativity, less predictable

Temperature ranging from 0 to 1

Temperature in the model call

# Conservative generation
response1 = client.messages.create(
    model="claude-sonnet-4-20250514",
    temperature=0.2,
    messages=[{"role": "user", "content": prompt}])

# Creative generation
response2 = client.messages.create(
    temperature=0.8,
    messages=[{"role": "user", "content": prompt}])

print("Temperature 0.2 Response (Conservative)")
print(response1.content[0].text)

print("Temperature 0.8 Response (Creative)")
print(response2.content[0].text)

Temperature in the model call

Temperature 0.2 Response (Conservative)
A rainbow is an optical phenomenon that occurs when sunlight and rain are present 
simultaneously. It forms when white light from the sun passes through water 
droplets in the atmosphere, causing the light to refract, reflect, and disperse 
into its component colors...

Temperature 0.8 Response (Creative)
Ah, rainbows! Those magical bridges painted across stormy skies that never fail 
to make us pause and wonder. Picture this: millions of tiny water droplets hanging 
in the air after a rainfall, each one acting like a miniature crystal ball. When 
sunlight hits these droplets, something absolutely enchanting happens...

Using top_p parameter

top_p (nucleus sampling): Alternative to temperature for creativity control
Typical values: 0.1 (focused) to 0.9 (diverse)
Combined use: Can use temperature and top_p together for precise control

client.messages.create(
model="claude-sonnet-4-20250514",
temperature=0.7,
top_p=0.3, #Likely words
messages=[{"role":"user", 
           "content":prompt}])

Claude's safety features

📝 Content strengths: Writing, analysis, coding, creative tasks
🚫 Built-in safety: Refuses harmful content, maintains ethical guidelines
✅ Best practices: Clear prompts, experiment with parameters, review outputs

Let's practice!

Introduction to Claude Models