🤖
STUDENT NOTES · DAY 17 OF 24
How AI Actually Works 🤖
ChatGPT feels like magic. Today you learn the trick — and it's simpler (and cooler) than you think.
1The one big trick: guess the next word
🐍
Py: An LLM plays one game, billions of times: "given this text, what word probably comes next?" Guess, add it, guess again. That's it. That's the whole magic.
Play the LLM! What comes next?
Once upon a ___
The model gives every word a probability — then usually picks a likely one.
2Words for your new world
🗣️ Prompt — the text YOU send in. 🧩 Token — a chunk of text (≈ part of a word) — models read and bill in tokens. 🌡️ Temperature — low = safe and predictable, high = creative and wild.
🦉
Professor Hoot: Better prompt → better answer. "Explain loops" is okay; "Explain loops to a 10-year-old using cricket" is chef's kiss. 🤌
3Talking to an LLM from Python
🤖
Bittu: Same requests.get idea as the weather API — different URL, smarter replies. You already have this skill!
4The warning label ⚠️
🐞
Bugly: LLMs sometimes state WRONG things very confidently. It's called hallucination — my proudest collaboration.
🐍
Py: They generate plausible text, not verified facts. Always double-check numbers, names and citations. Trust, but verify!
5Quick quiz! 🧠
At its core, an LLM…
💡 Predict → append → repeat, at massive scale. Simple trick, astonishing results.
Turning temperature UP makes answers…
💡 Low temp = safe librarian. High temp = wild poet. Choose per task!
6Your mission 🚀
Mission 1: Ask the same question to the class AI at temperature 0 and 1.5 — compare the answers.
Mission 2: Craft a prompt that explains dictionaries using a cricket analogy. Refine it twice.
Mission 3: Catch the AI hallucinating: ask about a fake event and watch it improvise. Screenshot it!
Day 17 done — the magician's trick is yours! 🎩
Next: build your own AI chatbot with memory.
Go to Day 18: Build a Chatbot →