Why what's in the fridge matters more than the recipe
Every time I need to explain how a large language model works with context, I think of my mother's pizza.
I'm five or six years old, we're living in Dallas, and the pizza looks like something that would send any Neapolitan into hysterics at first glance. It was more like an open-faced pie with tall edges — scorching hot. While it baked, the kitchen filled with a smell so thick and mind-blowing your mouth watered when you were just standing there.
I loved watching my mother get started. She had no set recipe, no careful shopping list, only a method that worked. She'd just open the fridge and lay whatever she could find in there out on the table:
Half a roast chicken, left over from last night's dinner.
The remains of some unidentifiable tomato sauce.
A couple of strips of bacon.
A handful of olives.
Some suspicious-looking sausages.
A wilted bunch of greens.
Dried-out cheddar and some other cheese.
And so on.
She'd take a good look at everything and say, "I wonder, Nicky, what we're going to come up with today." Then she'd pull the flour from the cabinet, grab the bottle of oil, start kneading dough, and shoo me out the door so I wouldn't get underfoot.
The base lineup of ingredients changed every time. The chicken might give way to a turkey leg or pork ribs, sausage might stand in for the links, olives might get swapped for artichokes. The only things that never showed up on that table were peanut butter and blueberry jam. We were making pizza, after all.
Whatever went into the filling, the ending was always the same: incredible. I never learned the actual secret to making it, since I was always kicked out of the kitchen and only called back once the work was done. Maybe it was magic.
You're not alone. Jennings didn't know what to do with all this either
It doesn't matter what you're sitting down to do — a presentation, a quarterly report, an in-depth piece, or an article about language models. The hardest part is always the start. You've got a rough sense of what you want to say. You've got a pile of raw material: a couple of ideas scribbled on a sheet of paper, chunks of old drafts saved somewhere, a file with source tables sitting off to the side, even a few quotes. There you are, staring at this pile of riches, with absolutely no idea what to do with it. Your drafts don't feel like ingredients for the perfect dish. They feel like meaningless trash.
Right about now, you're feeling exactly what Michael Jennings felt in Philip K. Dick's story "Paycheck." After two years working for a secretive corporation, he showed up for his payout — and instead of money, got an envelope full of strange junk:
A code key. A ticket stub. A parcel receipt. A length of fine wire. Half a poker chip, broken across. A green strip of cloth. A bus token.
"This, instead of fifty thousand credits," he murmured. "Two years…"
— Philip K. Dick, "Paycheck"
You're staring at the scraps of your own thoughts the exact same way Jennings gawked at his "payment," with no idea how to string them together, what to grab onto first, or even where to begin.
Good manners get in the way of getting things done
Why does this happen? Because we were trained to suffer this way. As kids, we were taught to show up with a fully-formed question already in hand. Don't waste other people's time: figure it out first, then ask. That's a good rule, and it's a habit that often helps when dealing with other people, like clients or academic advisors. When working with the model, though, it gets in the way.
A second mechanism also comes into play: your brain reads mental clutter as a sign you're not ready to move on to the next concrete step. So you wait for your thoughts to somehow arrange themselves into order. They usually don't rush.
Jennings spent more than half the story trying to figure out why he'd been given a piece of wire and a broken poker chip, and what to do with them. The model will sort out your problem faster than that.
When you’re not sure what to do with your resources, that’s exactly the moment to open the fridge.
Open the fridge
Stop waiting.
Open a new chat. Write, in a single sentence, exactly what you need on the other end: an article, a presentation, a report, or an email. The format doesn't matter.
Dump everything you've got into the model. Put in drafts, margin notes, links that seemed important at the time, and other people's paragraphs that you saved for unclear reasons. Add spreadsheets full of numbers you don't fully understand and a voice memo you dictated to yourself at three in the morning.
Don't organize this combination. Don't try to tidy it up. Don't try to draw logical connections — if you could see them, you’d already have written the piece. Above all, don't apologize to the machine for the mess.
Leave the peanut butter and blueberry jam in the fridge. Dump everything else right out on the table.
Ask one question: "What can we do with all of this?"
The model will start cooking.
What's happening in the kitchen while we're away
The model doesn't think the way you do. It never waits around for inspiration, isn't remotely afraid of a blank page, and works with whatever you hand it. You can shove your material under its nose right away, or you can spend three days dawdling, picking your nose, running out for coffee, and making the LLM drag god-knows-what off the internet instead of just handing over the perfectly good context you already have.
The moment you give the machine your drafts, it starts pulling them apart, building out the logic, and solving your actual problem. That changes everything.
The more raw material you lay out on the table, the finer and more precise the connections the AI draws. It won't write you a perfect, finished text, but it will put together a structure or offer you three to pick from. It'll explain which pieces connect and which ones are pulling in different directions. It'll ask you questions you hadn't even gotten to yet.
That's when the paralysis lifts. You'll know exactly where to start.
I still don't know exactly what happens in the kitchen.
Maybe it's magic.
