Lecture Analysis: Large Language Models & Prompt Architecture

From Classical Logic to Probabilistic Generation

This AI Subsystem module of Week 7 marks the profound paradigm shift in computer science from classical deterministic execution to probabilistic machine learning algorithms and Large Language Models (LLMs).

• Deterministic Systems (C / Python Logic): In traditional systems like speller, every structural execution step is strictly hardcoded via discrete data structures (hash tables, tries) to yield 100% deterministic output.
• Large Language Models (LLM Mechanics): Grounded in deep neural networks trained across immense textual corpora, LLMs operate primarily via probabilistic mathematical predictions of the next optimal token (Next Token Prediction).

Prompt Architecture & System Constraints

• Prompt Engineering: Structuring input parameters systematically to guide the neural model, constricting the mathematical probability landscape to extract highly accurate and pertinent outputs.
• System Instructions (System Prompts): As demonstrated in the lecture file chat3.py, system instructions enforce rigorous architectural and persona guardrails on the model, granting developers ultimate control over AI behavior before processing end-user input.

===================================================================================
                   LLM SYSTEM PROMPT EXECUTION FLOW
===================================================================================

  [ System Instructions ] ──┐
                            ├──> [ LLM Neural Engine (GPT-5) ] ──> [ Filtered Token Output ]
  [ User Raw Prompt ]    ──┘

===================================================================================