Read every book.
Never once left the room.
Every essay she wrote got marked by a human reader — that’s how she learned what a good answer looks like.
She can recite anything. She’s never verified a thing with her own eyes.
Ask her a question, and she doesn’t walk to a book. She rebuilds the answer from everything she’s ever read — word by word, in the moment.
Usually she’s right. The pattern holds.
Sometimes she fills a gap with something that sounds right and isn’t.
She can’t tell the difference. Neither can you.
When she fills the gap with something that isn’t true.
And she says it with the same confidence as everything else.
Hallucination is all LLMs do.
…dream machines.
…we direct their dreams with prompts.
…looks like a bug, but it’s just the LLM doing what it always does.
Someone else. Your data. Yourself.
Hong Kong · Feb 2024
$25.6 million.
15 wire transfers. One video call.
US nonprofit · 2013
S.D.N.Y. · 2023
Fined $5,000.
Ordered to apologize to every judge named.
Role · Objective · Limits · Examples · Steps.
You don’t need all five every time. Start with Role + Objective. Add the rest as you go.
You don’t need all five every time. Start with Role + Objective. Add the rest as you go.
AI drafts. You decide.