This is the S4 warm-up homework — no submission required.
Purpose: turn S4's two constitutional slogans — "My Model, My Choice" and "My Data, My Control" — from words you heard in class into two conversations you've seen with your own eyes. The same question, run through two different scripts, produces answers that differ — but exactly where they differ isn't something you can learn by listening. You have to open two windows and compare them yourself.
30 minutes. One cup of coffee's worth of time, and you'll feel exactly which cut S5's "refine" needs to make.
Open WukongDojo, and create a new agent under your account.
① Click New Agent · give it a name (e.g. "Little Wukong A")
② Paste your S3 script in, whole, as the system prompt
③ Pick a model
④ Paste in the same real scenario you used in your S3 script as the user input
⑤ Get the answer → copy the whole thing
① Click New Agent again · give it a name (e.g. "Little Wukong B")
② Paste your S3 script in, then make **two edits** —
change one Iron Law / cut a redundant paragraph / add a line
defining your audience / change the tone — any of these count
③ Pick the **same model** (identical to Agent A)
④ Ask the same question again
⑤ Copy the whole answer
⚔️ Hard constraint: the two agents differ only in script. Same model, same question. You're comparing how script design changes output — not comparing models.
Put Agent A's and Agent B's answers side by side and look at them honestly across these three axes:
| Axis | Ask yourself |
|---|---|
| 🧠 Depth of reasoning | Who went one step further (verified an assumption first / broke the problem down first / asked a clarifying question first)? |
| 🔍 Attention to detail | Who noticed the unspoken hook buried in the user's input? |
| 🎭 Your flavor | Which one sounds more like the character your script wrote? Which one lets the "default ChatGPT personality" leak through? |
Fill this in for yourself (not submitted, but strongly recommended — otherwise you'll forget it in 30 seconds):
🌪 What differed in reasoning depth: ________________________
🪡 What differed in the details: ________________________
🍵 Which one I prefer: ______ (Agent A / Agent B)
Why: ______________________________
Meaning: your script has strong control over the agent. Adding or removing a single Iron Law is enough to change the whole flavor of the answer.
Think about it: when you refine in S5, test after every single line you change — your script has already reached the precision where one cut changes one outcome.
Meaning: your script may be too generic — the model is filling in the blanks with its own defaults, and your instructions aren't strong enough yet to override the model's default behavior.
Think about it: in S5 you need to write instructions that are more specific and sharper, so the model has no choice but to follow your path instead of improvising its own.
Meaning: your original S3 script had redundancy or contradiction in it — cutting it made the model more focused on the right thing.
Think about it: this is the most valuable outcome. The core of S5's refinement isn't "adding more" — it's cutting down to only what must stay, so every line has a reason to exist.
"My Model, My Choice · My Data, My Control — the first rule of WukongDojo."
This exercise isn't about "feeling how magical AI is" — it's about putting your hand on it yourself: how much control does the script you wrote actually have over the output? Once you see it clearly, S5's refinement has a real target to aim at.
No submission required, but strongly recommended — after you've compared them, write yourself one sentence:
The single move most worth protecting in this agent is: __________
Next week (Homework 2), I'm going to **write it explicitly** into
the script.
—— That one sentence is your starting point for S5.