· A quiet Christmas Day
On Christmas Day 2022, the useful trick was asking for a table
🤖 AI-assisted, human-curated. How we use AI →
Christmas Day 2022 doesn't have an AI story. Nobody shipped anything, and the papers and press releases waited for the new year. ChatGPT itself was 25 days old (it came out on November 30, according to Wikipedia), and most of the people who'd tried it were busy with family, food, or both.
So, the kitchen.
The paragraph problem
Take the version of this day plenty of households know. Eleven people, one oven, a turkey that needs three hours, two sides that need 400 degrees, a pie that needs 350, and a cousin who just texted that she's bringing a casserole that "only needs a quick warm-up."
Ask a chatbot to help with that and you'll usually get a friendly paragraph. "Start by putting the turkey in early, then while it rests you can bake the sides, and don't forget to..." It's all correct and almost useless, because nobody standing at a stove can read a paragraph. You need to see the whole day at a glance.
That's the whole lesson for today. When the answer is a plan, a comparison, or anything you'll look at more than once, ask for a table.
Name the columns
Asking for "a table" helps. Naming the columns helps a lot more, because the columns are where you tell the model what you actually care about.
Before:
Help me plan cooking for Christmas dinner. Turkey, roasted potatoes,
green beans, rolls, apple pie, and my cousin's casserole.
Dinner is at 5.After:
Make a table for Christmas dinner at 5 p.m. Columns: time,
what goes in or comes out of the oven, oven temperature, and
who's doing it (me, Sam, or Lee). One oven only. Dishes: turkey
(3 hours at 325), roasted potatoes (45 min at 400), green beans
(stovetop, 10 min), rolls (12 min at 375), apple pie (bake ahead,
warm 15 min), cousin's casserole (warm 20 min).
Flag any time two things need the oven at different temperatures.The second prompt does three things the first one doesn't. It gives the real numbers, so the model isn't guessing cook times. It names the columns, so the answer comes back shaped like a schedule. And it asks the model to point out conflicts, which is the thing you'd actually miss at 3:40 with flour on your hands.
R.Solace writes about prompts the way a shop owner talks about orders: what you ask for is what you get.