Gen AI — LLMs

What are LLMs?

Large Language Models are neural networks trained on vast text data to understand and generate human-like text.

How they work

1. Training: Learn patterns from billions of text samples
2. Tokenization: Break text into tokens
3. Prediction: Predict next token in sequence
4. Generation: Chain predictions to create output

Major LLMs

Model	Company	Strengths
GPT-4	OpenAI	Reasoning, coding
Claude	Anthropic	Safety, analysis
Gemini	Google	Multi-modal
LLaMA	Meta	Open source
Mistral	Mistral AI	Efficiency

Parameters

• GPT-3.5: 175 billion parameters
• GPT-4: Estimated 1.7 trillion parameters
• Claude 3: Unknown, very large
• LLaMA 2: 7B, 13B, 70B parameters

Temperature

# Low temperature: more focused, deterministic
response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Write a poem"}],
    temperature=0.3
)

# High temperature: more creative, random
response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Write a poem"}],
    temperature=0.9
)

Token limits

# Check token count
import tiktoken

encoder = tiktoken.encoding_for_model("gpt-4")
tokens = encoder.encode("Hello, world!")
print(len(tokens))  # 2 tokens

Mini Practice

1. Compare responses at different temperatures
2. Count tokens in a prompt
3. Test different models
4. Understand context windows

Up Next

Continue with Prompt Engineering - Crafting effective prompts.