Large Language Models (LLMs)

# Using an LLM (OpenAI GPT API Example)
import openai
# Set API key
openai.api_key = "your-api-key"
# Simple completion
response = openai.ChatCompletion.create(
    model="gpt-4",
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Explain what an LLM is in simple terms."}
    ],
    temperature=0.7,      # Control randomness
    max_tokens=150,       # Limit response length
    top_p=0.9            # Nucleus sampling
)
print(response.choices[0].message.content)
# Output: "A Large Language Model (LLM) is like a very smart computer
# program that has read millions of books and websites. It learned patterns
# in how humans write and communicate, so it can understand questions and
# generate human-like text responses. Think of it as an AI that can chat,
# write, translate, and help with various language tasks..."
# The model predicted this response token by token, each time choosing
# the most likely next word based on its training!

# Demonstrating Autoregressive Generation
"""
How LLMs generate text step by step:
USER PROMPT: "Write a haiku about AI"
GENERATION PROCESS (token by token):
Step 1:  "" → predict → "Silicon"
Step 2:  "Silicon" → predict → " minds"
Step 3:  "Silicon minds" → predict → " awakening"
Step 4:  "Silicon minds awakening" → predict → "\n"
Step 5:  "Silicon minds awakening\n" → predict → "Patterns"
Step 6:  "Silicon minds awakening\nPatterns" → predict → " in"
Step 7:  "...Patterns in" → predict → " the"
Step 8:  "...Patterns in the" → predict → " data"
Step 9:  "...Patterns in the data" → predict → "\n"
Step 10: "...\n" → predict → "Learning"
Step 11: "...Learning" → predict → " never"
Step 12: "...Learning never" → predict → " stops"
Step 13: "...Learning never stops" → predict → "<END>"
FINAL OUTPUT:
"Silicon minds awakening
Patterns in the data
Learning never stops"
KEY INSIGHTS:
- Each token is predicted based on ALL previous tokens (context)
- Model doesn't "know" the full poem in advance
- Probability distribution at each step (could sample different tokens)
- This is why setting random seed affects output!
- Computational complexity: O(n) sequential steps for n tokens
"""
# Simplified pseudo-code
def generate_text(prompt, max_tokens=50):
    context = tokenize(prompt)
    generated = []
    for _ in range(max_tokens):
        # Model computes probability distribution over vocabulary
        logits = model.forward(context)  # [vocab_size]
        probs = softmax(logits)
        # Sample next token (various strategies possible)
        next_token = sample(probs, temperature=0.7)
        generated.append(next_token)
        context.append(next_token)  # Add to context
        if next_token == END_TOKEN:
            break
    return detokenize(generated)
# This happens internally when you call:
# response = openai.ChatCompletion.create(...)

# Advanced Prompt Engineering Techniques
# 1. ZERO-SHOT (Direct instruction)
prompt_zero = """
Translate to French: "Hello, how are you?"
"""
# 2. FEW-SHOT (Provide examples)
prompt_few = """
Translate to French:
English: "Good morning"
French: "Bonjour"
English: "Thank you"
French: "Merci"
English: "Hello, how are you?"
French:
"""
# 3. CHAIN-OF-THOUGHT (Step-by-step reasoning)
prompt_cot = """
Question: Roger has 5 tennis balls. He buys 2 more cans of tennis balls. 
Each can has 3 tennis balls. How many tennis balls does he have now?
Answer: Let's think step by step.
1. Roger starts with 5 balls
2. He buys 2 cans, each with 3 balls
3. 2 cans × 3 balls/can = 6 balls
4. Total: 5 + 6 = 11 balls
Answer: 11 balls
Question: Janet has 4 apples. She gives 2 to her friend and buys 5 more. 
How many apples does she have?
Answer: Let's think step by step.
"""
# 4. INSTRUCTION FOLLOWING (Clear structure)
prompt_instruction = """
You are a helpful assistant that answers questions concisely.
Rules:
- Keep answers under 50 words
- Use simple language
- Cite sources when possible
Question: What is machine learning?
Answer:
"""
# 5. SYSTEM + USER PATTERN (ChatGPT style)
messages = [
    {
        "role": "system",
        "content": "You are an expert Python tutor. Explain concepts clearly with examples."
    },
    {
        "role": "user",
        "content": "What is a list comprehension?"
    }
]
# TEMPERATURE CONTROL
# Low temperature (0.0-0.3): Deterministic, focused, factual
response_factual = openai.ChatCompletion.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "What is 2+2?"}],
    temperature=0.1  # Almost deterministic
)
# High temperature (0.8-1.5): Creative, diverse, random
response_creative = openai.ChatCompletion.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Write a creative story opening."}],
    temperature=1.2  # More random/creative
)
# BEST PRACTICES:
# - Be specific and clear
# - Provide context and examples
# - Use delimiters (""", ---, ###) to separate sections
# - Specify output format (JSON, bullet points, etc.)
# - Iterate and refine based on results