In Episode 2, Lt. Col. David Paddock is joined by Capt. Adam Knapp to pull back the curtain on how large language models function; from the raw data they're trained on, to the reward mechanisms that shape their behavior, to the inference stage where they generate responses. The episode breaks down four key capabilities that make modern LLMs so powerful: system prompts, chain-of-thought reasoning, retrieval-augmented generation (RAG), and tool access. Hallucinations, reward hacking, and training biases are addressed honestly, along with practical guidance on how to use these tools with appropriate skepticism.