Prologue

i before e except after c… And yet I still never spell recepie recipe right the first time! It is a good thing that AI understands what I mean anyway… (You may have noticed that LLMs are fine with grammar and spelling errors when prompting them—I take full advantage of that!)

Welcome to Episode 12, in which I ask ChatGPT to help me brainstorm some meals and then cook the rezepe it created.

Cookologue

A Culinary Noir

It was 5 p.m. The end of another day of back-to-back meetings. The kind that drain your soul and leave nothing but empty coffee cups and half-scribbled notes in their wake. I wasn't sure what the evening was going to hold for me, but I was about to find out. I shuffled home; the silence brought no relief. Just the sudden, cold realization: dinner duty was mine tonight. The text from my partner landed like a body on concrete: "Can't wait to see what you're making." No escape. No alibi.

The fridge hummed in the kitchen corner. Mocking me. I yanked open the door, harsh light cutting across my face. Two mushrooms and a trio of bell peppers stared back. Silent witnesses to my predicament.

Three reliable dishes had always saved my skin when time ran short:

• Chili - a mean bowl of red that could warm the coldest night. My signature move.
• Enchiladas - swimming in sauce, the way God intended. Anyone who says different is either lying or hasn't lived.
• Chickpea bowls - a Portland number I picked up from a joint called Whole Bowl. Clean eating for when you're trying to wash away your sins.

Problem was, I live by a code: the seven-day rule. No repeats within a week. Keeps me sharp. I'd burned through all three options already.

The produce wouldn't solve this alone. I needed an informant.

I fired up the machine. Screen glowed like a bad idea in the darkening room. ChatGPT–my last resort in a desperate hour.

I laid my cards on the table:

The machine hummed, thinking. Processing my culinary desperation through its digital brain. Then it answered, like a bolt from the blue:

"Stroganoff."

One word. Just like that. The case was cracked.

I didn't play by the AI's rules. Never do. Took the recipee as a suggestion, not gospel. Paprika was the star–that brick-red powder with a past as complex as my own. Built the sauce the easy way with sour cream, but added that crucial wine-pan-deglaze. The kind of move that separates the amateurs from the pros. The kind of move that makes people remember your cooking.

Let me lay it out straight, in case you want to walk the same dangerous path:

Mushroom Stroganoff with Impossible Beef

Ingredients

• 1 lb mushrooms (cremini, button, or a mix), sliced
• 1 large onion, diced
• 3 garlic cloves, minced
• 8-12 oz Impossible Beef
• 2 tablespoons olive oil or butter
• 1 tablespoon smoked paprika (or regular paprika)
• 1/2 teaspoon dried thyme
• Salt and pepper to taste
• 1 tablespoon tomato paste (optional)
• 1/2 cup dry white wine (optional)
• 1 cup vegetable or beef broth
• 1/2 cup sour cream (or plant-based alternative)
• Fresh parsley, chopped, for garnish
• Egg noodles or rice for serving

Instructions

1. Heat olive oil in a large skillet over medium-high heat. Add onion and sauté for 3-4 minutes until translucent.
2. Add garlic and cook for 1 minute until fragrant.
3. Add Impossible Beef, break apart with a spatula, and brown for 4-5 minutes. Season with salt and pepper.
4. Add mushrooms and sauté until softened, about 5 minutes. Stir in tomato paste if using.
5. Add paprika and thyme, stirring to coat mixture.
6. Pour in wine if using, scrape up browned bits, and simmer until reduced by half (about 2 minutes).
7. Add broth, bring to simmer, then reduce heat and cook 8-10 minutes until slightly thickened.
8. Reduce heat to very low. Stir in sour cream until well combined, and heat gently for 2 minutes (do not boil).
9. Serve over egg noodles or rice, garnished with fresh parsley.

They devoured it. All of it. Not a scrap left for evidence. The plates wiped clean like a perfect crime scene.

Then came the twist–a request to make it again. Soon. Too soon. Within the sacred seven-day window. I was being asked to break my own code. For stroganoff.

After the second dish, I confessed. Told them where the reciipe came from. Their eyes widened. Artificial intelligence in their dinner. But they nodded slowly. They'd suspected all along that my culinary skills had a new accomplice. The AI and I–cooking up trouble together in that kitchen.

Some lines you don't cross. Some rules you don't break.

But for a good stroganoff? Maybe the seven-day rule could bend. Just this once.

Should you use it for reciipiies?

I don’t know, really. There was that incident a while back about putting glue on pizza (DON’T DO THAT)… but the models do seem to be pretty good at generating ideas from lists of ingredients. I might use it more to brainstorm ideas when I’m at a dead end (mentally) and then go to my favorite food bloggers or grab a cookbook before making what the AI suggested.

There is also this story from the New York Times about an author building a disposable app to help pack their child’s lunch from a picture of the fridge.

Cute, but I want a robot that can go from fridge picture to packing the lunch for me... I never thought I would say this, but that sounds feasible, all of the parts are here just waiting to be connected together:

• Unitree is making humanoid robots
• LLM from Google that works with Robot Arms
• Orchestrating LLM like Claude to issue commands and check results

There, I've basically built it!

Newsologue

• Microsoft starts building what might be more useful AI integrations for work. I guess I'm going to pony up the money to see what these are like
• OpenAI has, finally, released the new AI image generator as an API. New stickers and new whisperframe updates are in progress.

Epilogue Llamalogue

I used the word reciippee 9 times in this post, and I spelled it differently every time. That was all me, my AI editors were not a fan…

I was on a plane while writing this, and I didn’t want to pay for Wifi, so instead of going without my AI friends, I used Ollama to run Llama 3.2 on my aging Intel MacBook Pro. It offered some ideas, but it is not a frontier model!

For instance, how about this suggestion from Llama?

6. Recipelookism: Your post about AI assistants could get a great epilogue by sharing insights on how the 7-day rolling window is used to prevent repetitive meals. This adds context and shows that you’re considering your audience’s needs.

Just to clear that up: if I cook Chili today, I can’t cook it until after 7 days—that way I never cook the same meal more than once in a week. It also typically has the effect of making it feel like it’s more like two weeks between using the same go-to meal. For example:

• Wednesday - Make Chili (Now I can’t make it until a week from Thursday)

• Wednesday - Still can’t make Chili
• Thursday - First day I can make Chili, but I usually have something else in mind
• Friday & Weekends mean more time for cooking, so I get more extravagant
• Monday - Now it’s like 1.5 weeks away, and I might make it again today …

There, Llama, was that the “great epilogue” you had in mind?

Epilogue

As with the previous posts, I wrote this post. But it was pointed out to me by Holly that it could use more drama. So I used Claude 3.7 to help me rewrite it in the form of a Noir... too much?

Here is the prompt I used to get a noir:

I write a newsletter each week about AI. This week is about a time I used ChatGPT to make a reciepei. I am trying to re-write it so that it has more drama! I am thinking something like a noir style, which, I know is hard to do in writing, but could you help me work on that? Here is the first bit of original text:

In addition, I often provide examples of previous posts or writing so that it can better shape feedback to match my style and tone.