Under the Hood – The Forensic Mechanics of LLMs

Lesson 1.3: Vectors & Dimensions – How AI “Sees” Ideas

Lesson Content

When you type a prompt into ChatGPT, Claude, or Gemini, it feels like someone is sitting on the other side thinking up an answer. It feels like magic.

But as an AI Architect, your first job is to kill the magic.

An LLM does not “think.” It does not “know” things. It is a highly advanced, ultra-fast predictive typing engine. Think of it as the autocomplete on your smartphone, but scaled up to look through billions of pages of text.

.Probability vs. Intelligence

Instead of understanding concepts, an LLM breaks language down into small chunks called tokens (usually a word or a few characters). When you give it a prompt, the machine calculates a mathematical probability string to guess what token should come next.

Mathematically, it looks at the history of the words you provided and calculates:

If you type: “The detective walked into the room and found a dead…” The machine runs the numbers based on everything it has read on the internet. It doesn’t see a crime scene; it sees percentages:

• body (85% probability)

• phone (10% probability)

• end (5% probability)