Session 01 · Concept
How an LLM actually works — why it sometimes lies and sometimes feels like magic
90 min
Topics
- 01.1What an LLM is — a next-word predictor trained on patterns, not a database. Why that changes how to ask.
- 01.2Semantic space — why “luxury condo” and “premium oceanfront unit” sit near each other, and why word choice quietly changes results.
- 01.3Hallucinations — when the model invents data with full confidence, and where that’s a liability in real estate (comps, legal, pricing, HOA).
- 01.4Context window — what the model forgets, and how to structure long buyer conversations so you stay in frame across weeks.
How it’s taught
Concepts paired with live FL demos: a hallucination demo on a made-up Brickell address, a semantic-space test with rookie vs. seasoned vocabulary, and a context experiment on a six-week out-of-state buyer thread.
You leave with
By the end, the Realtor stops thinking of AI as “Google on steroids” and starts treating it as a language probability model — the single shift that changes how they prompt and where they trust it on a deal.
Between sessions · Write — by hand — a 100-word explanation of why AI hallucinates, aimed at a skeptical seller who just asked you to “let ChatGPT price the house.” Read aloud at the start of class 2.