How Language Models Work

Large Language Models (LLMs) like ChatGPT, Claude, and others are the technology behind modern AI assistants. Understanding how they work - at a conceptual level - helps you use them more effectively and recognize their limitations. The key insight: they predict what text should come next, one piece at a time.

Objectives

• Understand the basic concept of next-token prediction
• Explain why LLMs can seem knowledgeable without "knowing" anything
• Recognize how training data shapes model behavior
• Understand why LLMs can generate plausible-sounding false information
• Appreciate the difference between retrieval and generation

Key Takeaways

• LLMs predict text one piece at a time based on learned patterns
• They do not retrieve facts from a database - they generate likely text
• Training data determines what patterns the model learned
• Plausible-sounding output is not the same as correct output
• The model has no way to verify what it generates